Changes in CT radiomic features predict survival and early response to CDK 4/6 inhibitors in hormone receptor positive metastatic breast cancer

doi:10.21203/rs.3.rs-2002969/v1

Download PDF

Article

Changes in CT radiomic features predict survival and early response to CDK 4/6 inhibitors in hormone receptor positive metastatic breast cancer

https://doi.org/10.21203/rs.3.rs-2002969/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 11 Aug, 2023

Read the published version in npj Breast Cancer →

You are reading this latest preprint version

Background

The combination of Cyclin-dependent kinase 4/6 inhibitors (CDKi) and endocrine therapy (ET) is the standard of care for hormone receptor-positive (HR+), human epidermal growth factor receptor 2-negative (HER2-) metastatic breast cancer (MBC). Currently, there are no robust biomarkers that can predict response to CDKi, and so far, it is not clear which patients benefit from this therapy. Because MBC patients with liver metastases have a poorer prognosis and are consequently often treated with chemotherapy rather than endocrine therapy, developing predictive biomarkers that could identify patients likely to respond to CDKi is clinically important. We hypothesized that changes (“delta”) in the radiomic texture patterns on CT scans both within and outside metastatic liver lesions before and after CDKi therapy are associated with overall survival (OS) and can accurately assess early response to CDKi therapy.

Methods

From a retrospective patient registry, 73 HR + MBC patients with known liver metastases who received palbociclib (palbo) plus ET were identified from two different academic sites. One site was used as a training set (n = 32), while the other as a validation set (n = 41). Patients with objective response/stable disease per RECIST v1.1 were defined as ‘responders’, and those with progressive disease within 6 months were labelled ‘non-responders’. Radiomic texture and shape features measuring subtle differences in lesion heterogeneity and size on a pixel level were extracted from pre-treatment and post-treatment CT scans within the lesions measured for RECIST assessment, and the difference (delta) radiomic features were computed. Delta radiomic features were selected by least absolute shrinkage and selection operator with the Cox regression model within the training set and top selected features along with their corresponding coefficients were used for radiomic risk score (RRS) construction. The RRS was further evaluated for association with OS within the validation set. Patients were stratified into low and high-risk groups based on an ideal threshold of RRS identified in the training set and the association of RRS with OS was assessed with a log-rank test, Hazard ratio (HR (95% CI), and Harrell’s concordance index (C-index). In addition, a linear discriminant analysis (LDA) classifier was trained with identified features to predict RECIST-derived response in the validation set.

Results

RRS was found to be significantly associated with OS in training (HR: 2.9; 95% CI, 1.6–5.5; P = 0.0006; C-index = 0.82) and validation sets (HR: 2.4; 95% CI, 1.06–5.6; P = 0.035; C-index = 0.77). Median OS times in high and low-risk groups were 12.58 and 23.17 months, respectively (P = 5.7e-04). Compared to RECIST response in the training and validation set, delta radiomic features were able to assess early response with a ROC curve AUC of 0.74 and 0.72, respectively.

Conclusions

Delta radiomics analysis can reasonably predict response and survival in HR+/HER2- MBC patients treated with CDKi in combination with endocrine therapy.

Endocrine therapy (ET) is the preferred treatment for patients with hormone receptor-positive (HR+), and human epidermal growth factor receptor 2-negative (HER2-) metastatic breast cancer (MBC) [1]. However, about half of the patients who receive ET will eventually develop therapeutic resistance, and subsequently derive limited clinical benefit [2].

More recently, therapies inhibiting cyclin-dependent kinase 4/6 have been demonstrated to delay the development of therapeutic resistance when combined with ET [1] and consequently significantly improve progression-free survival (PFS) compared to ET alone among patients with advanced HR+, HER2-negative breast cancer [3–5]. CDK4/6 inhibitors (CDKi) effectively block the cell cycle proliferation from G1 (pre-DNA synthesis) to the S phase (DNA synthesis) by blocking the CDK4/6-cyclin D1 complex and preventing cancer cell proliferation and treatment resistance [6].

However, the objective response rate (ORR) to CDKi plus ET is approximately 51–59% for first-line therapy and 20–48% for second line therapy [7]. Unfortunately, all patients including those initially sensitive to CDKi will eventually develop resistance [8, 9].

As a result, to date, there is no robust and reliable biomarker that can predict which HR+, HER2- MBC patients will clinically benefit from CDKi therapy, i.e., greater than six months OS [10]. Consequently, there is an unmet clinical need for novel biomarkers that can rationally guide the use of CDKi to identify patients most likely to benefit from treatment, avoid time on ineffective medications, as well as financial toxicities and potential adverse effects (neutropenia, diarrhea, transaminase elevation, diarrhea) in those unlikely to respond. This will allow oncologists to adjust treatment options early on and develop more successful therapeutic strategies to overcome endocrine resistance in CDKi non-responders [11].

Recently, computerized feature analysis of radiographic scans or radiomic analysis has demonstrated significant potential for prediction of response to chemo- and targeted therapy in breast cancer [12, 13]. These radiomic approaches can computationally capture quantitative measurements of tumor heterogeneity and its microenvironment in radiological images, such as computed tomography (CT). In the metastatic setting, patients usually undergo serial CT scans throughout treatment to monitor disease progression. Recent evidence suggests that microarchitectural and subtle morphologic changes within the tumor might precede changes in gross radiographic features (tumor size and shape) in response to treatment [14].

In this study, we aimed to use a delta radiomic-based analysis of breast cancer patients with liver metastases on pre- and post-treatment CT to predict early treatment response to CDKi therapy. We hypothesized that quantitative capture of textural changes of the lesion and the surrounding microenvironment in the CT scan before and after treatment may provide a better accurate characterization of treatment response compared to the textural pattern on CT scan before initiating therapy. Towards this end, we used CT scans from 73 MBC patients with liver metastasis at baseline- and post-treatment with CDKi. We sought to identify radiomic features associated with RECIST response and overall survival (OS) in advanced breast cancer patients treated with CDKi by interrogating the tumor and tumor microenvironment on CT imaging.

Study population

This multi-institutional study included HR+ metastatic breast cancer (MBC) patients with liver metastasis, who received palbociclib (palbo), ribociclib (ribo), or abemaciclib (abema) as first- or second-line therapy in combination with ET.

A chart review was performed to identify patients with HR+ MBC, with liver metastasis and available baseline and post-treatment CT abdomen/pelvis at University Hospitals/Seidman Cancer Center (UHSCC, n=52) and Cleveland Clinic (CCF, n=45).

Scans of patients not suitable for feature extraction, such as those with CT scan artifacts and poor image quality, absence of post-treatment scans, or non-contrast CTs were excluded. This resulted in a total of n=32 patients from UHSCC and n=41 patients from CCF. UHSCC was used for training (S_t) and CCF was used as an independent validation cohort (S_v).

The study conformed to Health Insurance Portability and Accountability Act (HIPAA) guidelines and was approved by the Institutional Review Board (IRB) at University Hospitals (STUDY20201206) and Cleveland Clinic (IRB 19-559). The IRB waived the requirements for patient informed consent due to the retrospective, and observational nature of this study.

Clinical endpoints

The primary endpoint of this study was overall survival (OS). OS was measured from the date of CDKi initiation, until either a recorded death or censored at the last known date of follow-up. The secondary endpoint was radiographic response status, as defined by RECIST v1.1. Patients who had progressive disease after CDKi therapy were classified as ‘non-responders’ and patients who had a complete response, partial response, or stable disease were classified as ‘responders’.

Lesion segmentation and feature extraction

Baseline and post-treatment CTs were acquired either on Siemens, GE Medical Systems, Philips, or Toshiba scanners according to standard scanning protocol at CCF and UHSCC institutions. The imaging protocol included a tube voltage of 100 to 120 kVp, slice thickness ranging from 1 to 3 mm, and standard convolution kernel reconstruction. All patients were injected with a contrast injection before imaging. An expert reader reviewed and annotated liver lesions on baseline and post-treatment scans using a freehand tool on 3D-Slicer® software. The annotations were verified by a board-certified radiologist (14 years of experience). The primary lesion – noted in the radiology report and corresponding to the largest size post-treatment treatment CT – was chosen as the region of interest for radiomic analysis. In patients where liver lesions could not be confidently defined on CT scans, a recent dedicated liver MRI was utilized for confirmation of metastasis. Radiomic texture features were extracted on a pixel level from pre-treatment and post-treatment CT scans within the lesions measured for RECIST assessment. For each scan, a set of intra- and peritumoral radiomic features were considered. The peritumoral rim around the lesion was defined via the use of quantitative morphological operations (dilation) as a region extending radially from the lesion boundary up to roughly 12 mm around the tumor. The choice of peritumoral compartment rim size was defined based on the previously described method [15], including the normal appearing liver parenchyma surrounding the metastatic lesion and excluding regions containing peri-hepatic fat (fat has much lower attenuation compared to normal liver parenchyma at CT, with a range of −10 to −100 HU) [16].

Radiomic texture features (consisting of 13 Haralick texture features [17], 25 Laws features [18], 25 Laws Laplacian, and 48 Gabor features [19]) were extracted from 2D contours, in a slice-by-slice manner across all annotated slices of the lesion, on both pre-treatment and post-treatment scans. These features capture textural patterns, heterogeneity, and local appearance of the tumor and its microenvironment on CT and reflect hallmarks of tumor biology. Each texture feature was summarized by five first-order statistics (mean, median, SD, skewness, kurtosis) separately within the tumor and peritumoral region, resulting in a total of 555 texture feature statistics. Moreover, 24 shape features that capture aspects of the 3D lesion structure including the size, volume, and longest diameter of the tumor were also extracted. All radiomic feature values were then normalized (mean=0 and SD=1) and the change of feature statistics between baseline and post-treatment scans was calculated to yield the feature set. The overall experimental design for this study is shown in Figure 1.

Statistical Analysis

The least absolute shrinkage and selection operator (LASSO) method was used to select the most prognostic features to OS in S_t [20]. The top selected features along with their corresponding coefficients were used for radiomic risk score (RRS) construction. RRS was calculated via a linear combination of non-zero coefficient selected features weighted by their respective coefficients. The value of the tuning parameter (penalty parameter) in the LASSO model (λ) was averaged out by 100 cross-validations to minimize error. The LASSO Cox regression model was performed using the “glmnet” package in R.

A risk score threshold was chosen in S_t to stratify patients into high and low-risk groups based on the median of RRS. The prognostic performance of RRS was validated using Kaplan-Meier survival analysis, log-rank test, HR (95% CI), and Harrell's concordance index (C index [95% CI]). Univariate analysis of RRS and the clinical–pathological variables was performed to evaluate the association of each marker with OS. Multivariable Cox regression analysis was used to investigate the independent prognostic value of the RRS relative to clinical-pathological variables.

A prognostic nomogram was also developed on S_t by combining clinical and prognostic radiomic features and validated on S_v. To evaluate nomogram performance, C indices were calculated from the nomogram for RRS, clinical factors alone, and RRS plus clinical factors. The calibration plot for the nomogram was evaluated by reviewing the plots of nomogram-predicted survival probabilities with Kaplan-Meier estimated probabilities along with the Hosmer-Lemeshow test, a statistical test for goodness of fit for logistic regression models. A p-value < 0.05 implies that the model is not a good fit whereas the converse suggests that there is no evidence of poor fit. Bootstraps with 500 resamples were employed to quantify model overfitting and for calculating Kaplan-Meier estimates. The nomogram model was generated by the use of R with the “rms” package (Regression Modeling Strategies). A decision curve analysis (DCA) was performed to evaluate the clinical efficacy of the radiomics model by assessing the net benefits of the model across a range of threshold probabilities [21].

Additionally, a linear discriminant analysis (LDA) was used to evaluate the ability of identified prognostic features to predict early response to therapy. The LDA classifier generates linear class boundaries (i.e., linear patterns) while assuming that the covariance of each class is identical. The classifier performance for predicting response was evaluated by the area under the receiver operating characteristic (ROC) curve (AUC). In S_t, the classifier performance was assessed by averaging the AUC values computed over 100 iterations of threefold cross-validation (CV). The classifier was eventually trained on S_t and tested for response prediction within S_v.

Finally, any differences between clinical categories were assessed using Fisher’s exact test, where a two-sided t-test was used for continuous variables. A bilateral P-value < 0.05 was considered statistically significant.

Of the 32 patients in S_t, 65% of patients had objective response/stable disease on the first imaging assessment on CDKi/ET as per RECIST v1.1, and 35% had progressive disease. The median age at diagnosis was 63 years [35–82]. 21 of them were White, 4 were African American, and race information was unavailable for the remaining 7 patients. All patients were ER + and 9% were also HER2+. 65% of patients received palbociclib as 1st or 2nd line therapy and the remaining received a different CDKi, i.e., Ribociclib or Abemaciclib. After initiating ET/CDKi therapy, 24/32 (75%) of the patients had a progression of the disease. The median time from the start date of CDKi to the date of progression was 12 months (95% CI, 10.32–18.3).

Of the 41 patients in S_v, in the first radiographic assessment, 22 had an objective response and 19 had progressive disease. The median age at diagnosis was 58 [36–79] years. Out of the 41 patients, 16 were White, 3 were African American, and the self-reported race information for the remaining patients was not available.

The median overall survival for the patients in S_t was 18.15 months (95% CI, 11.26–25) and for the patients in S_v was 19.43 months (95% CI, 14.93–23.93). A univariable Cox regression analysis identified that OS did not significantly differ for: race (White vs. African American) (HR: 0.85 (95% CI, 0.22–3.26); P = 0.81; C-index = 0.54), age (HR: 1 (95% CI, 0.95–1.04); P = 0.99; C-index = 0.48), or tumor volume (before CDKi/ET therapy) (HR: 1.2; 95% CI, 0.71–2; P = 0.49; C-index = 0.56).

Delta Radiomic features from pre- and post-treatment CT scans were associated with OS in patients treated with CDKi

Within S_t, 7 radiomic features were obtained from 1110 radiomic features after feature pruning from the LASSO model. The LASSO model selected 7 radiomic features with a lambda value of 0.18 (see Fig. 2A). Details of the selected features and their coefficients have been listed in Fig. 2B. Of the 7 radiomic features, 3 were picked from the peritumoral region and 4 were selected from the intra-tumoral region. The RRS ranged from − 1.58 to 2.87 and the optimum cut-off value (the median) was found to be -0.119. Based on this value, patients were divided into high- and low-risk groups. A univariable Cox regression analysis developed using radiomic features indicated that RRS was significantly associated with OS in S_t (HR: 2.9 (95% CI, 1.6–5.5); P = 0.0006; C-index = 0.82) and S_v (HR: 2.4 (95% CI, 1.06–5.6); P = 0.035; C-index = 0.77). Median survival time in high and low-risk groups was 12.58 and 23.17 months, respectively (P = 5.7e-04). In a multivariable analysis using a combination of clinical and radiomic features, the RRS alone was found to be significantly associated with OS in S_t (risk-score: HR = 2.65 (95% CI: 1.47–4.8), P = 0.0012; age: HR = 1, 95% CI: 0.95–1.06, P = 0.8; race: HR = 0.56, 95% CI: 0.07–4.2, P = 0.57; baseline tumor vol: HR = 1.8 (95% CI: 0.72–4.4), P = 0.2; C-index = 0.82) and S_v (risk-score: HR = 2.4 (95% CI: 1.02–5.6), P = 0.044; tumor vol: HR = 1.2 (95% CI: 0.78–1.86), P = 0.4; C-index = 0.78). The corresponding Kaplan–Meier survival curves showed a significant difference in OS between patients with low and high RRS both in S_t and S_v (P < 0.05). Kaplan–Meier survival curves for S_t and S_v are shown in Fig. 2C and 2D, respectively.

A radiomics nomogram model incorporating the radiomics signature with clinical biomarkers was the model that best predicted OS with a C-index of 0.82 (95% CI, 0.73–0.91) in S_t compared to the clinical or radiomics model alone (Fig. 3A). In S_v, the C-index was 0.79 (95% CI, 0.71–0.86). The clinical model alone had a lower prognostic performance (compared to radiomics alone) in predicting OS with a C-index of 0.57 (95% CI, 0.40–0.75) in S_t.

The calibration plot (Fig. 3B) demonstrated a good fit between nomogram-predicted and observed OS. The Hosmer-Lemeshow test yielded a p-value of 0.47, suggesting no significant difference between predicted and observed OS. The DCA was used to demonstrate the clinical decision utility of the nomogram. Figure 3C shows DCA for three models (clinical model, radiomic model, and integrated clinical plus radiomics model). The integrated model had the highest net benefit in the prediction of high-risk patients (those with poor OS) to receive more intensive treatment (e.g., chemotherapy) than the clinical model or radiomics model alone.

Tumor volumetric changes during CDKi therapy were not associated with OS

The median tumor volume in S_t was 2.57 mL (range, 0.23–33.32 mL) before CDKi administration and 3.9 mL (range, 0.121–20.84 mL) after therapy. The tumor change size during therapy was not statistically associated with OS, neither in S_t (HR: 0.82 (95% CI, 0.43 − 1.58); P = 0.56; C-index = 0.46), nor in S_v (HR: 0.7 (95% CI, 0.34 − 1.42); P = 0.32; C-index = 0.49).

Delta radiomic features assess early response to ET/CDKi therapy

Figures 4A and 4B illustrate the discriminability of the intranodular Haralick entropy delta radiomic feature for representative non-responder and responder patients before and after CDKi therapy. We observed an elevated expression of Haralick entropy post-therapy in the non-responders as compared to the responders. Moreover, the LDA classifier trained with identified prognostic features yielded an AUC of 0.74 (95% CI, 0.61–0.98) on S_t. Prediction accuracy was slightly different between patients with thin slice thickness (1 mm) and thick slice thickness (3 and 5 mm), but it was not statistically significant (P > 0.05). Within S_v, the classifier yielded an AUC of 0.72, with an accuracy of 0.7, sensitivity of 0.67, and specificity of 0.86.

The use of CDKi with ET has revolutionized the management of HR + HER2- metastatic breast cancer due to their favorable toxicity profiles and their relevant antitumor activity [5, 22, 23]. The US Food and Drug Administration (FDA) and European Medicines Agency (EMA) have approved the clinical use of CDKi such as palbociclib, ribociclib, and abemaciclib in metastatic breast cancer patients with HR+/HER2- in the first line setting as well as the second line in combination with ET. In clinical trials, all FDA approved CDKi when combined with ET have demonstrated a significant prolongation of PFS compared with ET alone [5, 6].

Approximately half of all patients with metastatic breast cancer develop liver metastases and 5–12% of patients exhibit liver metastases as the primary site of breast cancer recurrence [24, 25]. If untreated, liver metastases are associated with a dismal prognosis and a median survival of only 4–8 months [26]. Moreover, most MBC patients who respond to first-line therapy will progress within 1–2 years [27], and 5-year OS rates for all MBC subtypes are less than 25%. In addition, ER + MBC patients with lung or liver metastasis compared to patients with the bone predominant disease have a significantly shorter median PFS to ET +/- CDKi [28, 29]. Hence, there is an urgent need for clinically useful predictive biomarkers that can identify patients with liver metastases likely to benefit from ET/CDKi.

Many previous studies have investigated the role of molecular alterations in the tumor such as pRB, RB1 mutations, CCDN1amplifications, and CCNE1 overexpression as potential biomarkers for CDKi response but these investigations have not yielded any identification of a clinically useful predictive biomarker so far [30–32]. A growing body of research suggests that loss of the retinoblastoma tumor suppressor gene (Rb), leads to accelerated angiogenesis and tumor progression which is one of the most important biomarkers associated with acquired resistance and lower PFS to CDKi [33, 34].

In this study, we investigated the role of delta radiomic features on baseline and post-treatment CT scans from HR+/HER2- breast cancer patients with liver metastasis to predict response and overall survival after CDKi therapy. A nomogram model that integrated radiomic scores with clinical biomarkers was developed in this study. Our nomogram model showed that radiomic scores had a better prognostic performance for predicting OS compared to clinical biomarkers alone. Moreover, the decision curve analysis (DCA) showed that the radiomic score had a better overall net benefit compared to clinical biomarkers for predicting high-risk patients suitable to receive more aggressive therapy across several threshold probability values. To the best of our knowledge, this work is the first attempt for predicting response in breast cancer patients treated with CDKi in the context of radiomic feature analysis.

We found that higher intratumoral Haralick entropy that captures tumor heterogeneity was associated with non-response to CDKi/ET and poor OS. Previous studies have shown that tumor heterogeneity increment is indicative of genomic heterogeneity and is associated with a worse prognosis in non-small cell lung cancer patients treated with immunotherapy or chemotherapy [35–37]. By contrast, decrement of intratumoral heterogeneity is associated with a favorable response to therapy and prolonged PFS. A previous study by Wander et al. showed that harmonic alterations in RB1, AURKA, and CCNE2 expression enhance sensitivity to CDKi therapy. In other words, heterogeneity increase resistance to CDKi therapy [34].

Angiogenesis is another hallmark of cancer proliferation and tumor metastases [38]. We found that peritumoral Laws feature increment after therapy, likely reflecting increased angiogenesis and tumor microenvironment heterogeneity, which is associated with poor therapeutic response and OS [33, 34, 39].

In addition, p16 (tumor suppressor) downregulation leads to increased HIF-α which in turn causes tumor hypoxia, which is known to confer therapeutic resistance [40]. Prior evidence suggests that a hypoxic tumor environment might be captured by radiomic texture analysis of lesions extracted from CT images [35].

Moreover, there is also some evidence suggesting that CDKi can elicit their therapeutic response by enhancing the activation of T-cells [41]. It can be postulated that increment in peritumoral Gabor radiomics after therapy might be capturing the increment in tumor immune environment that is caused by CDKi therapy.

Finally, we found that delta radiomic features were also able to assess early response to CDKi/ET. Our results suggest that our radiomic model can help to monitor early response in patients undergoing CDKi therapy. The ability to determine early response during treatment will allow early adjustment of treatment regimens. In the future, such validated image-based radiomic biomarkers can potentially identify non-responders and will enable oncologists to predict residual endocrine sensitivity and reduce ineffective treatment, toxicity, and side effects associated with CDKi therapy and timely change to other effective target therapies, including subsequent CDK4/6 and PI3K/AKT/mTOR inhibitors [42]. Such validated biomarkers can also identify those patients that would benefit from CDKi versus those patients that would benefit from the standard status quo.

We acknowledge that our study did have its limitations. While we used two independent cohorts of patients for building and validating our model, the cohort size in this study is relatively small but it is quite challenging to assemble large cohorts for this problem due to a small number of patients treated with this relatively new therapy. The second limitation is the retrospective nature of our study, not a prospective study. To tackle this limitation, validation on a large multi-site prospective cohort is required. Also, there are questions on variability in scanning differences between scanners such as convolution kernels, reconstruction algorithms, and slice thickness, that hinder widespread applicability of radiomic features, although some studies have shown novel radiomic features that are relatively immune to differences in image-related variabilities [43]. Also, further work is needed to be done to perform extensive stratified analyses to explore the relationship between the molecular and mutational status of the tumors and radiomics in this patient population.

We hope to address these limitations in future works. In addition, we need to develop and validate the signature as predictive of the maximum benefit of CDKi therapy. However, it needs a randomized controlled trial (RCT) design (where patients were treated with two regimens - status quo and CDKi) and then demonstrates the predictive benefit of CDKi therapy.

Nonetheless, despite these limitations, our study revealed that dynamic change of CT-based radiomic texture features between baseline and post-treatment of HR+, HER2- breast cancer patients with liver metastasis can predict early response and OS to CDKi coupled with ET.

Competing interests: NB is a current employee of Picture Health and a former employee of Tempus Labs and IBM Research, with whom he is an inventor on several pending patents pertaining to medical image analysis. He additionally holds equity in Picture Health and Tempus Labs. AM is an equity holder in Picture Health, Elucid Bioimaging, and Inspirata Inc. Currently, he serves on the advisory board of Picture Health, Aiforia Inc, and SimBioSys. He also currently consults for Biohme, SimBioSys and Castle Biosciences. He also has sponsored research agreements with AstraZeneca, Boehringer-Ingelheim, Eli-Lilly and Bristol Myers-Squibb. His technology has been licensed to Picture Health and Elucid Bioimaging. He is also involved in 3 different R01 grants with Inspirata Inc.

Authors’ contributions: Conceptualization, M.K., V.V., and A.M.; methodology, M.K., V.V. and A.M.; software, M.K.; validation, M.K., V.V. and A.M.; formal analysis, M.K.; investigation, M.K. and V.V. and N.B.; resources, M.K.; data curation, V.V., AJ.M.; writing—original draft preparation, M.K., V.V. and N.B.; writing, review and editing, M.K., V.V., N.B., AJ.M., and A.M.; visualization, M.K.; supervision, A.M.; All authors have read and agreed to the published version of the manuscript.

Acknowledgments: Research reported in this publication was supported by the National Cancer Institute under award numbers R01CA26820701A1, R01CA249992-01A1, R01CA202752-01A1, R01CA208236-01A1, R01CA216579-01A1, R01CA220581-01A1, R01CA257612-01A1, 1U01CA239055-01, 1U01CA248226-01, 1U54CA254566-01, National Heart, Lung, and Blood Institute 1R01HL15127701A1, R01HL15807101A1, National Institute of Biomedical Imaging and Bioengineering 1R43EB028736-01, National Center for Research Resources under award number 1 C06 RR12463-01, VA Merit Review Award IBX004121A from the United States Department of Veterans Affairs Biomedical Laboratory Research and Development Service the Office of the Assistant Secretary of Defense for Health Affairs, through the Breast Cancer Research Program (W81XWH-19-1-0668), the Prostate Cancer Research Program (W81XWH-15-1-0558, W81XWH-20-1-0851), the Lung Cancer Research Program (W81XWH-18-1-0440, W81XWH-20-1-0595), the Peer Reviewed Cancer Research Program (W81XWH-18-1-0404, W81XWH-21-1-0345, W81XWH-21-1-0160), the Kidney Precision Medicine Project (KPMP) Glue Grant and sponsored research agreements from Bristol Myers-Squibb, Boehringer-Ingelheim, Eli-Lilly, and AstraZeneca.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health, the U.S. Department of Veterans Affairs, the Department of Defense, or the United States Government.

Data Availability: Data are available upon reasonable request. Access to datasets from the Cleveland Clinic and the University Hospitals Cleveland Medical Center (used with permission for this study) should be requested directly from these institutions via their data access request forms. Subject to the institutional review boards’ ethical approval, unidentified data would be made available as a test subset. All experiments and implementation details are described thoroughly in the Materials and methods section so they can be independently replicated with non-proprietary libraries.

Preusser, M. et al. CDK4/6 inhibitors in the treatment of patients with breast cancer: summary of a multidisciplinary round-table discussion. ESMO Open 3, e000368 (2018).
Osborne, C. K. & Schiff, R. Mechanisms of Endocrine Resistance in Breast Cancer. Annu. Rev. Med. 62, 233–247 (2011).
Finn R S, Crown J, Lang I, Boer K, Bondarenko I, et al. Overall survival results from the randomized phase II study of palbociclib (P) in combination with letrozole (L) vs letrozole alone for frontline treatment of ER+/HER2– advanced breast cancer (PALOMA-1; TRIO-18). J. Clin. Oncol. 35, 1001–1001 (2017).
Im SA, Lu YS, Bardia A, et al. Overall Survival with Ribociclib plus Endocrine Therapy in Breast Cancer. N Engl J Med. 2019;381(4):307-316. doi:10.1056/NEJMoa1903765
Goetz MP, Toi M, Campone M, et al. MONARCH 3: Abemaciclib As Initial Therapy for Advanced Breast Cancer. J Clin Oncol. 2017;35(32):3638-3646. doi:10.1200/JCO.2017.75.6155
Preusser, Matthias et al. “CDK4/6 inhibitors in the treatment of patients with breast cancer: summary of a multidisciplinary round-table discussion.” ESMO open vol. 3,5 e000368. 20 Aug. 2018, doi:10.1136/esmoopen-2018-000368.
Shah, Mirat et al. “CDK4/6 Inhibitors: Game Changers in the Management of Hormone Receptor–Positive Advanced Breast Cancer?.” Oncology (Williston Park, N.Y.) vol. 32,5 (2018): 216-22.
Pandey, Kamal et al. “Molecular mechanisms of resistance to CDK4/6 inhibitors in breast cancer: A review.” International journal of cancer vol. 145,5 (2019): 1179-1188. doi:10.1002/ijc.32020.
Konecny GE, Winterhoff B, Kolarova T, et al. Expression of p16 and retinoblastoma determines response to CDK4/6 inhibition in ovarian cancer. Clin Cancer Res. 2011;17(6):1591-1602. doi:10.1158/1078-0432.CCR-10-2307.
Cardoso, F. et al. 3rd ESO–ESMO International Consensus Guidelines for Advanced Breast Cancer (ABC 3). Ann. Oncol. 28, 16–33 (2017).
Thill, M. & Schmidt, M. Management of adverse events during cyclin-dependent kinase 4/6 (CDK4/6) inhibitor-based treatment in breast cancer. Ther. Adv. Med. Oncol. 10, (2018).
Braman NM, Etesami M, Prasanna P, et al. Erratum to: Intratumoral and peritumoral radiomics for the pretreatment prediction of pathological complete response to neoadjuvant chemotherapy based on breast DCE-MRI. Breast Cancer Res. 2017;19(1):80. Published 2017 Jul 10. doi:10.1186/s13058-017-0862-1.
Braman N, Prasanna P, Whitney J, et al. Association of Peritumoral Radiomics With Tumor Biology and Pathologic Response to Preoperative Targeted Therapy for HER2 (ERBB2)–Positive Breast Cancer. JAMA Netw Open. 2019;2(4):e192561. doi:10.1001/jamanetworkopen.2019.2561.
Seymour L, Bogaerts J, Perrone A, Ford R, et al: guidelines for response criteria for use in trials testing immunotherapeutics. Lancet Oncol. 2017 Mar;18(3):e143-e152. doi: 10.1016/S1470-2045(17)30074-8. Epub 2017 Mar 2. Erratum in: Lancet Oncol. 2019 May;20(5):e242.
Terayama, Noboru; Matsui, Osamu; Ueda, Kazuhiko; et al. Peritumoral Rim Enhancement of Liver Metastasis: Hemodynamics Observed on Single-Level Dynamic CT During Hepatic Arteriography and Histopathologic Correlation, Journal of Computer Assisted Tomography: November 2002 - Volume 26 - Issue 6 - p 975-980.
Zhang YN, Fowler KJ, Hamilton G, Cui JY, et al. Liver fat imaging-a clinical overview of ultrasound, CT, and MR imaging. Br J Radiol. 2018 Sep;91(1089):20170959. doi: 10.1259/bjr.20170959.
Haralick RM, Shanmugam K, Dinstein I. Textural features for image classification. IEEE Trans Syst Man Cybern 1973;SMC-3:610–21.
Laws K. Textured image segmentation [dissertation]. Los Angeles (CA): University of Southern California; 1980.
Fogel I, Sagi D. Gabor filters as texture discriminator. Biol Cybern 1989;61:103–13.
Tibshirani R. The lasso method for variable selection in the Cox model. Stat Med 1997;16(4):385–395.
Vickers AJ, Elkin EB. Decision curve analysis: a novel method for evaluating prediction models. Med Decis Making 2006;26(6):565–574.
Finn RS, Martin M, Rugo HS, et al. Palbociclib and letrozole in advanced breast cancer. N Engl J Med. 2016;375(20):1925–1936. doi: 10.1056/NEJMoa1607303.
Rugo H, Finn R, Diéras V, et al. Palbociclib plus letrozole as first-line therapy in estrogen receptor-positive/human epidermal growth factor receptor 2-negative advanced breast cancer with extended follow-up. Breast Cancer Res Treat. 2019;174(3):719–729. doi: 10.1007/s10549-018-05125-4.
He ZY, Wu SG, Peng F, et al. Up-Regulation of RFC3 Promotes Triple Negative Breast Cancer Metastasis and is Associated With Poor Prognosis Via EMT [published correction appears in Transl Oncol. 2020 Jun;13(6):100803]. Transl Oncol. 2017;10(1):1-9. doi:10.1016/j.tranon.2016.10.004.
Bale R, Putzer D, Schullian P. Local Treatment of Breast Cancer Liver Metastasis. Cancers (Basel). 2019;11(9):1341. Published 2019 Sep 11. doi:10.3390/cancers11091341.
Adam R, Aloia T, Krissat J, et al. Is liver resection justified for patients with hepatic metastases from breast cancer?. Ann Surg. 2006;244(6):897-908. doi:10.1097/01.sla.0000246847.02058.1b.
Cristofanilli M, Hortobagyi GN. New horizons in treating metastatic disease. Clin Breast Cancer. 2001 Jan;1(4):276-87. doi: 10.3816/CBC.2001.n.002.
Robertson, J. F. R. et al. Fulvestrant 500 mg versus anastrozole 1 mg for hormone receptor-positive advanced breast cancer (FALCON): an international, randomised, double-blind, phase 3 trial. Lancet Lond. Engl. 388, 2997–3005 (2016).
Reddy P, Martin J, Montero A. Real World Data and Independent Predictors of Clinical Outcomes With CDK Inhibitors in Metastatic ER+ Breast Cancer Patients NCCN 2022 annual conference J Natl Compr Canc Netw 2022;20(5.5):551–560 abstract HSR22-163.
Turner NC, Liu Y, Zhu Z, Loi S,ColleoniM, Loibl S, et al. Cyclin E1 expression and palbociclib efficacy in previously treated hormone receptor-positive metastatic breast cancer. J Clin Oncol 2019;37:1169–78.
Ma CX, Gao F, Luo J, Northfelt DW, Goetz M, Forero A, et al. NeoPalAna: neoadjuvant palbociclib, a cyclin-dependent kinase 4/6 inhibitor, and anastrozole for clinical stage 2 or 3 estrogen receptor-positive breast cancer. Clin Cancer Res 2017;23:4055–65.
Anurag M, Haricharan S, Ellis MJ. CDK4/6 Inhibitor Biomarker Research: Are We Barking Up the Wrong Tree? Clin Cancer Res. 2020 Jan 1;26(1):3-5. doi: 10.1158/1078-0432.CCR-19-3119.
Wu, Yanmei et al. Current Therapeutic Progress of CDK4/6 Inhibitors in Breast Cancer. Cancer management and research vol. 12 3477-3487. 15 May. 2020, doi:10.2147/CMAR.S250632.
Wander SA, Cohen O, Gong X, Johnson GN, et al. The genomic landscape of intrinsic and acquired resistance to cyclin-dependent kinase 4/6 inhibitors in patients with hormone receptor-positive metastatic breast cancer. Cancer Discov. 10, 1174–1193 (2020). doi: 10.1158/2159-8290.CD-19-1390.
Khorrami M, Prasanna P, Gupta A, et al. Changes in CT Radiomic Features Associated with Lymphocyte Distribution Predict Overall Survival and Response to Immunotherapy in Non-Small Cell Lung Cancer. Cancer Immunol Res. 2020;8(1):108-119. doi:10.1158/2326-6066.CIR-19-0476.
Khorrami M, Khunger M, Zagouras A, et al. Combination of Peri- and Intratumoral Radiomic Features on Baseline CT Scans Predicts Response to Chemotherapy in Lung Adenocarcinoma. Radiol Artif Intell. 2019;1(2):e180012. Published 2019 Mar 20. doi:10.1148/ryai.2019180012.
Khorrami M, Jain P, Bera K, et al. Predicting pathologic response to neoadjuvant chemoradiation in resectable stage III non-small cell lung cancer patients using computed tomography radiomic features [published correction appears in Lung Cancer. 2019 Oct;136:156]. Lung Cancer. 2019;135:1-9. doi:10.1016/j.lungcan.2019.06.020.
Folkman J. What is the evidence that tumors are angiogenesis dependent?. J Natl Cancer Inst. 1990;82(1):4-6. doi:10.1093/jnci/82.1.4.
Hasina R, Pontier AL, Fekete MJ, et al. NOL7 is a nucleolar candidate tumor suppressor gene in cervical cancer that modulates the angiogenic phenotype. Oncogene. 2006;25(4):588-598. doi:10.1038/sj.onc.1209070.
Laitala A, Erler JT. Hypoxic Signalling in Tumour Stroma. Front Oncol. 2018;8:189. Published 2018 May 29. doi:10.3389/fonc.2018.00189.
Deng J, Wang ES, Jenkins RW, et al. CDK4/6 Inhibition Augments Antitumor Immunity by Enhancing T-cell Activation. Cancer Discov. 2018;8(2):216-233. doi:10.1158/2159-8290.CD-17-0915.
Presti D, Quaquarini E. The PI3K/AKT/mTOR and CDK4/6 Pathways in Endocrine Resistant HR+/HER2- Metastatic Breast Cancer: Biological Mechanisms and New Treatments. Cancers (Basel). 2019 Aug 24;11(9):1242. doi: 10.3390/cancers11091242.
Khorrami M, Bera K, Thawani R, et al. Distinguishing granulomas from adenocarcinomas by integrating stable and discriminating radiomic features on non-contrast computed tomography scans. Eur J Cancer. 2021;148:146-158. doi:10.1016/j.ejca.2021.02.008.

There is a conflict of interest NB is a current employee of Picture Health and a former employee of Tempus Labs and IBM Research, with whom he is an inventor on several pending patents pertaining to medical image analysis. He additionally holds equity in Picture Health and Tempus Labs. AM is an equity holder in Picture Health, Elucid Bioimaging, and Inspirata Inc. Currently, he serves on the advisory board of Picture Health, Aiforia Inc, and SimBioSys. He also currently consults for Biohme, SimBioSys and Castle Biosciences. He also has sponsored research agreements with AstraZeneca, Boehringer-Ingelheim, Eli-Lilly and Bristol Myers-Squibb. His technology has been licensed to Picture Health and Elucid Bioimaging. He is also involved in 3 different R01 grants with Inspirata Inc.

Download PDF

Journal Publication

published 11 Aug, 2023

Read the published version in npj Breast Cancer →

Editorial decision: revise
20 Oct, 2022
Review #2 received at journal
13 Oct, 2022
Reviewer #2 agreed at journal
29 Sep, 2022
Review #1 received at journal
25 Sep, 2022
Reviewer #1 agreed at journal
08 Sep, 2022
Reviewers invited by journal
01 Sep, 2022
Submission checks completed at journal
29 Aug, 2022
Editor assigned by journal
26 Aug, 2022
First submitted to journal
26 Aug, 2022

You are reading this latest preprint version

Changes in CT radiomic features predict survival and early response to CDK 4/6 inhibitors in hormone receptor positive metastatic breast cancer

Status:

Journal Publication

Version 1

Abstract

Background

Methods

Results

Conclusions

Figures

Introduction

Materials And Methods

Results

Discussion

Declarations

References

Additional Declarations

Status:

Journal Publication

Version 1