AI-based Model CRM to Evaluate the Responses of Breast Cancer Patients to CDK4/6 Inhibitors-Based Therapies and Simulate Real-World Clinical Trials
===================================================================================================================================================

* Mei Yang
* Yuhan Liu
* Yi-Ching Hsueh
* Qiangzu Zhang
* Yanhui Fan
* Juntao Xu
* Min Huang
* Xu Li
* Guangming Tan
* Chunming Zhang
* Jianfei Yang
* Gang Niu

## Abstract

Breast cancer patients exhibit diverse responses to CDK4/6 inhibitor (CDK4/6i)-based therapies, and identifying eligible patients remains a challenge. Artificial intelligence (AI) has demonstrated the potential to address complex clinical problems. Here, we applied a novel AI-based approach, named as CDK4/6i Response Model (CRM), which combined a previously published method and a scoring model based on random forest algorithm for evaluating breast cancer patients’ sensitivity to CDK4/6i-based therapies. To train the CRM, we transformed the genomic data of 980 breast cancer patients from the TCGA database into signaling pathway activity profiles (APSP) by utilizing the modified Damage Assessment of Genomic Mutations (DAGM) algorithm. To mimic the mechanism of action of CDK4/6 inhibitors, a scoring model was then trained to classify the HR+/HER2- and HR-/HER2-breast cancer molecular subtypes by the differential APSP features between the two, which reasonably reflected the potential role played by CDK4/6 molecules in HR+/HER2-breast cancer cells. The effectiveness of the CRM’s ability was verified by accurately classifying HR+/HER2- and HR-/HER2-breast cancer patients in a separate local patient cohort (n = 343) in Guangdong, China. Significantly, the scores were observed to be distinct (p = 0.025) between CDK4/6i-treated patients with different responses. Furthermore, breast cancer patients belonging to different subtypes were grouped into five distinct populations based on the scores assigned by the CRM. The results showed not only the heterogenetic responses across subtypes but also more than half of HR+/HER2+ patients might be benefited from CDK4/6i-based treatment. The CRM empowered us to conduct in-silico clinical trials (ICT) on different types of cancer patients responding to CDK4/6i-based therapies. In this study, we performed twin ICT of previously disclosed clinical trials ([NCT02246621](http://medrxiv.org/lookup/external-ref?link\_type=CLINTRIALGOV&access\_num=NCT02246621&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom), [NCT02079636](http://medrxiv.org/lookup/external-ref?link\_type=CLINTRIALGOV&access\_num=NCT02079636&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom), [NCT03155997](http://medrxiv.org/lookup/external-ref?link\_type=CLINTRIALGOV&access\_num=NCT03155997&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom), [NCT02513394](http://medrxiv.org/lookup/external-ref?link\_type=CLINTRIALGOV&access_num=NCT02513394&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom), [NCT02675231](http://medrxiv.org/lookup/external-ref?link_type=CLINTRIALGOV&access_num=NCT02675231&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom)), and observed concerted results as the real-world clinical outcomes. These findings show the potential of CRM as a companion diagnostic for CDK4/6i-based therapies and demonstrate promising applications by ICT to guide pan-cancer treatment using CDK4/6 inhibitors in the clinical ends.

**Summary** In most cases, the efficacy of CDK4/6 inhibitors (CDK4/6i) for breast cancer patients is tightly correlated with molecular subtypes, which are classified by the status of hormone receptors (HR) and human epidermal growth factor receptor 2 (HER2)1. More than one pathological condition in HR+/HER2-type of patients are approved to use CDK4/6i-based therapies as standard treatment2–10, while patients with HR-/HER2-breast cancer have less chances to benefit from CDK4/6i-based therapies11. However, there is currently no approach to differentiate the two types of breast cancer based on their differences in cellular functions, which leads to the vagueness in how they react differently to same kind of treatment including CDK4/6 inhibitors12,13. Furthermore, from clinical observation, partial HR+/HER2-patients are proved resistant to CDK4/6i14, and identifying eligible patients with all types of breast cancer remains a challenge15–17. Here we show an AI-based function explanatory method, named CDK4/6i Response Model (CRM), for evaluating breast cancer patients’ sensitivity to CDK4/6i-based therapies. The CRM trained by differential genomic data-transformed pathway-level features between HR+/HER2- and HR-/HER2-breast cancers was able to classify HR+/HER2- and HR-/HER2-breast cancer patients in TCGA (n=581) and an independent patient cohort (n=343), which also reflects the different responses to CDK4/6i-based therapies18. We found that the CRM scores are significantly different between CDK4/6i-treated patients with different outcomes. Breast cancer patients across different subtypes were mainly distributed into five populations regarding to the CRM scores, indicating not only the heterogenetic responses across subtypes but also more than half of HR+/HER2+ patients might be benefited from CDK4/6i-based treatment. The CRM empowered us to conduct in-silico clinical trials (ICT) which simulates previously disclosed clinical outcomes. Our ICT results reflected the efficacy differences observed clinically, which shows the potential of CRM to become a companion diagnostic for CDK4/6i-based therapies. Thus, our study demonstrated a novel and useful tool to explore therapeutic potentials of CDK4/6i-based treatments in breast cancer and, promisingly, other more malignant diseases.

## Introduction

In clinical practice, patients often respond differently to the same drug, making it challenging to identify the most suitable population for a particular treatment. Traditional methods for identifying biomarkers to predict drug sensitivity and resistance have been found ineffective, and the underlying mechanisms behind patient responses remain poorly understood19. As artificial intelligence (AI) methods continue to advance in biomedical research, they offer great potential for exploring the mechanisms of drug response20,21. Here, we present a study focused on CDK4/6 inhibitor (CDK4/6i)-based therapies, demonstrating the ability of AI methods to identify patients’ sensitivity to this treatment.

CDK4/6, cyclin dependent kinase 4 and 6, plays a key role in regulating the cell cycle of tumor cells22–24. Therefore, drugs targeting CDK4/6 have achieved success in some clinical applications, benefiting certain patient groups from CDK4/6 inhibitor (CDK4/6i)-based therapy1,25. For instance, CDK4/6 inhibitors have been approved by FDA for treating adult patients with hormone receptor (HR)-positive, human epidermal growth factor receptor 2 (HER2)-negative advanced or metastatic breast cancer in combination with endocrine therapy as initial treatment in postmenopausal women or in men14. Additionally, one of the CDK4/6i is also approved for decreasing the incidence of chemotherapy-induced myelosuppression in adult patients when administered prior to a platinum/etoposide-containing regimen or topotecan-containing regimen for extensive-stage small cell lung cancer26. Despite these successes, it is evident that the application conditions of CDK4/6i-based therapies are relatively narrow compared with other anti-cancer drugs. Currently, researchers are exploring further on CDK4/6 inhibitors in basic research and clinical aspects19,27–29. For instance, researchers are trying to determine the evolutionary stages of tumors that are more dependent on CDK4/6-related mechanisms to identify the appropriate timing for treatment30. Additionally, they are investigating how to develop CDK4/6-based multi-target drugs to enable more patients to benefit from complex clinical conditions31–33.

In the past, many explorations of biomarkers have been conducted to detect CDK4/6i-sensitive patients. However, the sensitivity evaluation method for CDK4/6i-based therapies is still an unmet clinical need, and no clear clinical practice standard can predict whether patients will respond to the treatment based on any single feature, like CCND1, CCNE1 and p16 loss15,16. Researchers also attempted to identify biomarkers for CDK4/6i resistance17. However, these biomarkers have not been effective in screening suitable patients in clinical settings. For example, only about 7% of patients were found to carry an RB1 gene loss-of-function mutation, which leads to congenital CDK4/6 inhibitor resistance4,34–36. Additionally, other explorations on single biomarkers, such as loss function of the FAT1 gene and amplification of FGFR1 and FGFR2, failed to screen suitable patients37,38. Multiple drug resistance mechanisms are involved in the CDK4/6 drug-resistant phenotype, and the population covered by these biomarkers is far less than the proportion of patients with actual drug resistance39,40. Thus, it is not yet possible to establish a clinically available patient screening method based on these biomarkers.

To date, despite numerous efforts to identify the characteristics of patients sensitive or resistant to CDK4/6 inhibitor (CDK4/6i)-based therapy, none of the findings have proven to be clinically useful35,41. The fundamental reason for this inefficiency in patient screening lies in the lack of a clear understanding of how tumors respond to CDK4/6i-based therapies, leading to inconsistent patient stratification based on response outcomes. The current sensitivity screening methods mainly rely on surface-level characteristics, such as single gene analysis, scattered single nucleotide polymorphisms (SNPs) of pharmacogenomics, and simple association analysis such as genome-wide association studies (GWAS)42–44. However, these methods fall short in revealing the underlying systemic mechanisms of drug sensitivity. Furthermore, these methods can be time-consuming and yield few tangible outcomes, as data accumulation over extended periods is often necessary, and even then, results may not be definitive.

Through clinical observations, it has been found that some therapies, such as endocrine therapy and CDK4/6 inhibitors, are effective for patients with HR+/HER2-breast cancer but not for those with HR-/HER2-breast cancer11,45. Therefore, it is promising to establish drug-sensitivity screening methods by comparing the systematic differences of HR+/HER2- and HR-/HER2-breast cancer in molecular biological characteristics to understand the underlying mechanism of drug sensitivity. Notably, CDK4/6 inhibitors exhibit different mechanisms of action on the two breast cancer subtypes, leading to significant differences in clinical efficacy. In HR+/HER2-breast cancer (including Luminal A and Luminal B1 molecular subtypes), CDK4/6i-based therapies directly inhibits tumor cell growth, resulting in significant improvements in patient outcomes46. However, the efficacy of CDK4/6i-based therapies for patients with triple-negative breast cancer (HR-/HER2-, TN molecular subtype) is relatively weak, with only a small number of patients benefiting from CDK4/6 inhibitors as a pre-treatment to protectively inhibit bone marrow-derived cells, ultimately leading to the continuation of immune system functions after other treatments47,48. Hence, it can be inferred that CDK4/6 inhibitors directly inhibit the growth of tumor cells, which is the common mechanism for a higher proportion of HR+/HER2-subtypes to benefit from CDK4/6i-based therapies.

An AI model capable of distinguishing between the HR+/HER2- and HR-/HER2-breast cancer subtypes based on their functional genomics differences could provide a general evaluation of patients’ suitability for CDK4/6i-based therapies. This would be an attractive way to establish a universal companion diagnostic method for better screening of sensitive patients. However, this idea is challenging, as currently there is no method, beyond immunohistochemical staining of HR and HER2 molecules, to differentiate between the two subtypes. Neither a simple variation distribution analysis of genome sequence nor complex machine learning methods have proven successful in this regard12,13.

In our previous study on the etiology of germline genomes in patients with various types of breast cancer, we proposed a potential solution49. By mapping germline rare coding variants (gRCVs) onto a quantitative set of signaling pathway profiles using the DAGM approach, we can easily model the functional patterns of cells driven by germline genomes in these patients. This method can not only distinguish between HER2-negative and positive patients but can also construct a scoring model to accurately predict the relative risk of HER2-negative breast cancer in female individuals, even those with wild-type BRCA1/2 and other potentially pathogenic genes. Building on this approach, we can modify the algorithm to analyze how somatic mutations drive deterministic changes in cell function, providing a methodological basis for distinguishing between different pathological tumor types.

In this study, we aimed to explore the potential of using tumor genomic data to establish a CDK4/6i response model (CRM), for distinguishing HR+/HER2- and HR-/HER2-breast cancer subtypes and assessing their sensitivity to CDK4/6i-based therapies. To achieve this goal, we analyzed the genome data of breast cancer patients in the TCGA database, which provided a comprehensive dataset of somatic mutations in cancer18. By converting these mutation data into clonal somatic activity profiles of signaling pathways (clonal sAPSP), we were able to capture the mechanistic features of tumor biology that underlie CDK4/6 inhibitor sensitivity. Through comparing the differences in clonal sAPSP between HR+/HER2- and HR-/HER2-breast cancer patients, we identified a set of informative features that could be used to distinguish the two subtypes. To establish a scoring model for patient classification, we employed a machine learning method, random forest, to combine the selected features and generate CRM for assessing patients’ sensitivity to CDK4/6i-based therapies50.

After verification of the CRM in patients treated with CDK4/6i-based therapies, we extended its application to other breast cancer subtypes to predict the efficacy of the therapy. However, due to limited availability of clinical trial data, it was challenging to directly validate the CRM using actual clinical data. Some efforts were made in conducting in-silico clinical trials (ICT) to overcome this kind of challenges51,52. In this study, we employed a computer simulation approach to establish digital twins of patients recruited in clinical trials based on TCGA data as ICT to show opportunities in presenting drug responses. This enabled us to simulate real clinical trials and evaluate whether the CRM can predict clinical outcomes. By this in-silico method, we provide a probable explanation of the observed clinical results and propose a potential companion diagnostic approach by CRM for CDK4/6i-based therapies, which could be a valuable clinical tool for selecting suitable patients for this kind of therapies.

## Results

### Development of CRM in HR+/HER2- and HR-/HER2-breast cancer patients

#### Subtype-specific characterization of TCGA-BRCA patients

In this study, TCGA breast cancer patients were classified into four subtypes based on their expression levels of ER, PR and HER2. All patients were calculated for clonal somatic activity profiles of signaling pathways (sAPSP) from genomic data using the adapted DAGM algorithm. The baseline characteristics of the patients were well-balanced across the subtypes (Table 1), with the majority of patients aged from 41 to 70 (72% in HR+/HER2- and HR+/HER2+, 93% in HR-/HER2+, 73% in HR-/HER2-) and most of them being female. The patient population in all subtypes included three or more races, such as Asian, black or African American, and white. Most of the patients were diagnosed at stage II according to staging information.

View this table:
[Table 1](http://medrxiv.org/content/early/2023/05/16/2023.05.15.23289976/T1)

Table 1 The baseline characteristics of the CRM training set and other TCGA-BRAC patients

### HR+/HER2- and HR-/HER2-subtypes exhibit the largest discrepancy in functional biological features

The clonal sAPSP of breast cancer patients were calculated for the total of the absolute Z score values across four subtypes. Results revealed that HR+/HER2- and HR-/HER2-subtypes had the largest difference in biological features, as indicated by the largest sum of absolute Z scores between the two subtypes. This suggests that these two subtypes exhibit the greatest discrepancy among breast cancer in functional biological features, and that the differential response to CDK4/6i-based therapies may reflect statistical differences in the mechanisms of different cancer subtypes.

#### Establishment of CRM for CDK4/6i-based therapies response prediction

The differential features between HR+/HER2- and HR-/HER2-subtypes were used to establish the CRM for predicting response to CDK4/6i-based therapies. APSP features with Z score >= 3 were selected, indicating significant variation between the two subtypes. For example, cell cycle G2M checkpoint regulation was excessively activated in HR-/HER2-subtype, while substantial inhibition appeared in HR+/HER2-subtype. In contrast, iCOS-iCOSL signaling in T helper cells was largely suppressed in HR-/HER2-breast cancer but stimulated in HR+/HER2-breast cancer.

The random forest method was used to establish the scoring model part of the CRM, which generates a score between 0 and 1. Patients with a score closer to 1 have tumor profiles closer to that of HR+/HER2-breast cancer patients and are suitable for CDK4/6i-based therapies. Conversely, patients with a score closer to 0 have tumor profiles closer to that of HR-/HER2-breast cancer patients and are likely not sensitive to CDK4/6i-based therapies. The CRM’s effectiveness was initially validated in the training set by the leave-one-out cross-validation (LOOCV) method, and the AUC for distinguishing HR+/HER2-from HR-/HER2-subtypes was 0.9956, as shown in Figure 1b.

![Fig.1](http://medrxiv.org/http://medrxiv.stage.highwire.org/content/medrxiv/early/2023/05/16/2023.05.15.23289976/F1.medium.gif)

[Fig.1](http://medrxiv.org/content/early/2023/05/16/2023.05.15.23289976/F1)

Fig.1 
The general concept of establishing the CRM to evaluate patients’ sensitivity to CDK4/6i-based therapies2–11

![Fig.2](http://medrxiv.org/http://medrxiv.stage.highwire.org/content/medrxiv/early/2023/05/16/2023.05.15.23289976/F2.medium.gif)

[Fig.2](http://medrxiv.org/content/early/2023/05/16/2023.05.15.23289976/F2)

Fig.2 The CRM for classification of HR+/HER2- and HR-/HER2-subtypes based on genomic data
**a**, Schematic of the model training, testing, and optimization workflow. **b**, DAGM analysis reveals significant differences in cell cycle G2M checkpoint regulation and iCOS-iCOSL signaling pathway between HR+/HER2- and HR-/HER2-subtypes in TCGA. **c**, ROC-AUC curves of the TCGA data-trained CRM for classifying TCGA (left) and local (right) HR+/HER2- and HR-/HER2-patients. **d**, ROC-AUC curves of the optimized CRM for classifying local HR+/HER2- and HR-/HER2-patients by leave-one-out cross validation (LOOCV). The accuracy curve demonstrates the model’s excellent separation ability. **e**, DAGM analysis of local patient dataset reveals significant differences in cell cycle G2M checkpoint regulation and iCOS-iCOSL signaling pathway between HR+/HER2- and HR-/HER2-subtypes. **f**, The CRM scores show a clear distinction between HR+/HER2- and HR-/HER2-subtypes in the local patient dataset.

#### CRM generalization ability confirmed in independent patient cohort

To evaluate the generalization ability of the CRM, an independent cohort of breast cancer patients from Guangdong was collected. The clonal sAPSP of 171 HR+/HER2- and 85 HR-/HER2-patients were calculated based on tumor genomic data and classified using the CRM. The results show that the CRM effectively distinguished between HR+/HER2- and HR-/HER2-subtypes (AUC=0.7563, accuracy=70% when cutoff set to 0.6, FDR=30%). These findings confirm the generalization ability of the model in a patient cohort of different ancestries and suggest its potential usefulness in clinical settings.

#### Optimization of CRM improves classification efficiency

After combining the TCGA and Guangdong local patient data for model training, the CRM’s effectiveness in distinguishing HR+/HER2- and HR-/HER2-patients was significantly improved during the LOOCV verification process. The AUC for classifying local patients increased from 0.7563 to 0.9795 after optimization, as shown in Figure 1d. This improvement in AUC indicates that the optimized CRM has a better ability to differentiate the two breast cancer subtypes from different ancestries. In addition, the accuracy of CRM to classify local patients was also improved from 70% to 95% when the cutoff was set to 0.6, and the FDR was reduced from 30% to 5%. These results suggest that the optimized CRM is more reliable and accurate in predicting the subtype of breast cancer patients, and thus can facilitate more personalized treatment decisions.

### The CRM scores effectively predict response to CDK4/6i-based therapies

In Guangdong patient cohort, based on the RECIST standard for solid tumors, 6 patients achieved partial response (PR), 6 were stable disease (SD), and 1 was disease progression (PD) after receiving CDK4/6i-based therapies. A total of 22 samples, including breast and lymphoid lesions, were analyzed using the CRM for these 13 patients. As shown in Figure 3A, the median CRM score of responding patients was 0.9181, significantly higher than that of non-responding patients (median: 0.8047, p=0.0209). The CRM scores therefore have a strong correlation with clinical outcomes when using CDK4/6i-based therapies.

![Fig.3](http://medrxiv.org/http://medrxiv.stage.highwire.org/content/medrxiv/early/2023/05/16/2023.05.15.23289976/F3.medium.gif)

[Fig.3](http://medrxiv.org/content/early/2023/05/16/2023.05.15.23289976/F3)

Fig.3 The CRM scores may reflect patient responses
**a**, Boxplots show the CRM scores of responders and non-responders to CDK4/6i-based therapies, indicating significantly higher scores in the responders group (p value=0.017). **b**, NMR imaging data before and after using CDK4/6i-based therapies illustrate a strong correlation between CRM scores and drug efficacy. **c**, Boxplots of the CRM scores for Luminal A and Luminal B1 subtypes of HR+/HER2-patients show significant variation. **d**, The HER2 positive subtypes exhibit a broader pattern than the Luminal B1 subtype, with partial overlap. **e**, Boxplot of HR-/HER2+ subtype scores indicate a lower score range compared with HR+/HER2+ subtype.

The CRM scores could retrospectively evaluate the responses of patients after treatment. In details, in a patient who had been treated with CDK4/6i-based therapies, the CRM score was 0.65, which indicated a low probability of response. As expected, the patient did not respond to therapy and showed SD after treatment. In contrast, for a patient with a CRM score of 0.95, who was treated with CDK4/6i-based therapies, the patient showed PR. These results demonstrate the ability of the CRM scores in evaluating the response of patients to CDK4/6i-based therapies and potential to be used as a prospective tool in the future in guiding clinical selection of appropriate therapy for breast cancer patients.

### Subtype-specific CRM score distribution suggests differential response to CDK4/6i-based therapies

The CRM scores of all collected breast cancer subtypes were analyzed to predict their response to CDK4/6i-based therapies. The CRM score describes the biological function of each tumor responding to CDK4/6i-based therapies at a certain point in the continuum (0 to 1) rather than dichotomy. It is noted that the score distributions of the five types of patients (Luminal A, Luminal B1, HR+/HER2+, HR-/HER2+, HR-/HR2-) vary significantly and also rank from high to low. Additionally, heterogeneity is observed in different types of breast cancer, shown as various peaks inside one distribution curve of a certain breast cancer subtype.

From CRM scores analysis, firstly, stratification of HR+/HER2-patients may be necessary when using CDK4/6i-based therapies. By combining the Ki-67 index, the highest scored HR+/HER2-patients were divided into Luminal A (Ki-67 Low) and Luminal B1 (Ki-67 High) types. Surprisingly, the CRM was able to identify the difference between Luminal A and Luminal B1 patients without incorporating the Ki-67 index in model training, suggesting that Luminal A and Luminal B1 types have distinct functional biology. As shown in Figure 3C, the scores of Luminal A type were significantly higher than those of Luminal B1 type. Both Luminal A and Luminal B1 subtypes displayed two peaks (peak 1: mean=0.95, peak 2: mean=0.73), with Luminal B1 subtype having more samples distributed in peak 2 than Luminal A subtype. These results suggest that Luminal A and Luminal B1 types should be evaluated separately when assessing the sensitivity of patients to CDK4/6i-based therapies.

In addition, by observation from the score distribution, CDK4/6i-based therapies may also benefit some patients with HR+/HER2+ and HR-/HER2+ tumors. The score distributions of HER2 positive patients show three distinct peaks (Figure 3D, peaks 1, 2, and 3: mean=0.40). The distributions of Luminal B1 and HER2 positive HR+/HER2+ patients have a significant overlapping region. However, HR+/HER2+ subtype primarily distributes in peaks 1 and 2, while HR-/HER2+ subtype allocates more in peaks 2 and 3 (Figure 3D). The HR-/HER2+ subtype also exhibits a mild peak 4 (mean=0.55), which is also present in the HR-/HER2-subtype (Figure 3E). These findings suggest that compared to HR+/HER2+ subtypes, only a small proportion of HR-/HER2+ patients may benefit from CDK4/6i-based therapies. It is essential to evaluate patients’ sensitivity to therapy accurately, taking into account the specific tumor subtypes.

Furthermore, the score comparison between HR-/HER2+ and HR-/HER2-subtypes indicates that a large proportion of HR-/HER2+ patients may be resistant to CDK4/6i-based therapies, as demonstrated by the large intersecting area of scores with HR-/HER2-subtype (Figure 3E). Based on the scores of HR-/HER2-subtype, three peaks are observed in the distribution, which are peak 2, peak 4, and peak 5 (mean=0.24). Peak 5 is a unique pattern of HR-/HER2-subtype that is not found in other breast cancer subtypes, indicating the uniqueness of HR-/HER2-subtype in responding to CDK4/6i-based therapies. Overall, the CRM score distribution of breast cancer subtypes could be a valuable predictor of response to CDK4/6i-based therapies.

### CRM for interpreting clinical outcomes and harnessing the potential by in-silico clinical trials

#### Insights from in-silico clinical trials on advanced-stage patients

CDK4/6 inhibitors have been approved for the treatment of HR+/HER2-advanced or metastatic breast cancer. To evaluate the efficacy of CDK4/6 inhibitors in this setting, we simulated the No.1 ([NCT02246621](http://medrxiv.org/lookup/external-ref?link_type=CLINTRIALGOV&access_num=NCT02246621&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom)) clinical trial, which examined the combination of CDK4/6i (Abemaciclib) and a nonsteroidal aromatase inhibitor in postmenopausal women with no prior systemic therapy in the advanced setting. The simulation of enrolled patients was based on the matched criteria listed in Table 2a. The clinical results showed that the combination therapy improved the ORR and PFS compared to the sole aromatase inhibitor arm (ORR: 55.4% vs 40.2%, PFS: 28.2 months vs 14.8 months) 9,53. CDK4/6i-based therapies is suitable for treating this group of patients.

View this table:
[Table 2](http://medrxiv.org/content/early/2023/05/16/2023.05.15.23289976/T2)

Table 2 The matched screening conditions for patients enrolled in simulated trials
**a**, for No.1 clinical trial ([NCT02246621](http://medrxiv.org/lookup/external-ref?link\_type=CLINTRIALGOV&access\_num=NCT02246621&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom)) and No.2 clinical trial ([NCT02079636](http://medrxiv.org/lookup/external-ref?link\_type=CLINTRIALGOV&access\_num=NCT02079636&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom)). **b,** for No.3 clinical trial ([NCT03155997](http://medrxiv.org/lookup/external-ref?link\_type=CLINTRIALGOV&access\_num=NCT03155997&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom)) and No.4 clinical trial ([NCT02513394](http://medrxiv.org/lookup/external-ref?link\_type=CLINTRIALGOV&access_num=NCT02513394&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom)). **c,** for No.5 clinical trial ([NCT02675231](http://medrxiv.org/lookup/external-ref?link_type=CLINTRIALGOV&access_num=NCT02675231&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom)).

In contrast, the efficacy of CDK4/6 inhibitors in stage IV non-small cell lung cancer (NSCLC) is limited. We simulated the No.2 ([NCT02079636](http://medrxiv.org/lookup/external-ref?link_type=CLINTRIALGOV&access_num=NCT02079636&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom)) trial, which compared different combinations of CDK4/6i with other drugs, like gemcitabine, ramucirumab and pemetrexed, to treat patients with stage IV NSCLC. All CDK4/6i combinations showed mere efficacy, with median PFS ranging from 1.58 to 5.55 months54. This suggests that CDK4/6i-based therapies is probably improper for patients with late-stage NSCLC.

Our simulation results show that the CRM scores and the efficacy of CDK4/6i-based therapies are tightly correlated when evaluating advanced-stage patients. As depicted in Figure 4B, the CRM scores of the No.1 trial evidence a significant discrepancy compared to the No.2 trial, reflecting the response differences in these two patient groups. These findings highlight the importance of personalized medicine and the potential of in-silico clinical trials to aid in treatment decision-making.

![Fig.4](http://medrxiv.org/http://medrxiv.stage.highwire.org/content/medrxiv/early/2023/05/16/2023.05.15.23289976/F4.medium.gif)

[Fig.4](http://medrxiv.org/content/early/2023/05/16/2023.05.15.23289976/F4)

Fig.4 Simulation of Clinical Trials
**a**, The workflow of the simulation process for clinical trials. **b**, Boxplots displaying significant differences in CRM scores between simulated No.1 and No.2 clinical trials. **c**, Score distributions of simulated No.3 and No.4 clinical trials exhibit notable divergence in the peak 2 region. **d**, Two distinct peaks were observed in the score distribution of the simulated No.5 clinical trial. A score threshold of 0.80 is set to differentiate the two peaks, resulting in 47% of patients being identified as potential candidates for CDK4/6i-based therapies.

### In-silico simulation reveals varying efficacy of CDK4/6i-based therapies in Luminal A and Luminal B1 subtypes

Based on Table 2b, the patients enrolled in the No.3 ([NCT03155997](http://medrxiv.org/lookup/external-ref?link\_type=CLINTRIALGOV&access\_num=NCT03155997&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom)) and No.4 ([NCT02513394](http://medrxiv.org/lookup/external-ref?link\_type=CLINTRIALGOV&access\_num=NCT02513394&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom)) trials are simulated, with well-balanced baseline characteristics between the two groups (Table 3). In line with our hypothesis, published clinical results confirm the association between CDK4/6 inhibitors and different outcomes in patients with Luminal A and Luminal B1 subtypes. Clinical trials were conducted to evaluate whether the addition of CDK4/6 inhibitors to standard adjuvant endocrine therapy would improve efficacy in early HR+/HER2-breast cancer patients. The primary outcome measured for these trials was invasive disease-free survival (iDFS). However, the results of the trials were divergent. For example, in the No.3 trial ([NCT03155997](http://medrxiv.org/lookup/external-ref?link\_type=CLINTRIALGOV&access_num=NCT03155997&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom), MONARCH E), the 3-year iDFS rate for the experimental and control arms was 88.6% and 82.9%, respectively, indicating that the CDK4/6 inhibitor improved the efficacy of adjuvant treatment55. In contrast, the No.4 trial ([NCT02513394](http://medrxiv.org/lookup/external-ref?link_type=CLINTRIALGOV&access_num=NCT02513394&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom), PALLAS) found almost identical 3-year iDFS rates for the experimental and control arms (89.4% vs. 89.3%), indicating that CDK4/6 inhibitors failed to provide additional benefits to standard endocrine therapy56. Although the No.4 trial achieved higher iDFS rates in both arms than the No.3 trial, the addition of CDK4/6 inhibitors did not improve the efficacy of standard endocrine therapy.

View this table:
[Table 3](http://medrxiv.org/content/early/2023/05/16/2023.05.15.23289976/T3)

Table 3 The baseline characteristics of the simulated No.3 and No.4 trials

Intriguingly, the Ki-67 index may play a crucial role in the varying outcomes of the two trials57. Specifically, the No.3 trial had a higher proportion of patients with Luminal B1 subtype breast cancer (57% with Ki-67 greater than 20%) 58. Furthermore, based on the No.3 trial results, CDK4/6i was approved for combination with adjuvant endocrine therapy in the treatment of Luminal B1 patients. Thus, the additional efficacy conferred by CDK4/6i is more pronounced in Luminal B1 subtype compared to Luminal A subtype, despite the better prognosis of Luminal A-type patients following treatment. This discrepancy may be attributed to the possibility that (1) Luminal A type patients have a higher degree of overlap in the response mechanisms of the two treatments, which might preclude the efficacy of CDK4/6 inhibitors; or (2) Luminal A type patients have different response mechanisms to the two treatments, but most of the benefited patients overlap.

As mentioned previously, Luminal A and Luminal B1 subtypes have similar two-peak distributions, but Luminal B1 samples tend to have a greater allocation at the minor peak compared to Luminal A samples. Following the steps of in-silico simulation (Figure 4A), we found that the distribution of the CRM scores, indicating responses to CDK4/6i-based therapies among HR+/HER2-patients, differs between the No.3 and No.4 clinical trials (Figure 4C, major peak: mean=0.95, minor peak: mean=0.70). The simulated No.4 trial has more samples distributed in the higher score range (major peak), while distribution of the No.3 trial leans towards the lower score range (minor peak). These results indicate better but indifferent efficacy in both arms of treatments in the No.4 trial. Also, it explains the lower but differential efficacy of the two arms of intervention shown in the No.3 trial. At this point, the CRM is capable for distinguishing the difference of responses to CDK4/6i-based therapies among HR+/HER2-patients. For HR+/HER2-patients at early stages, if the CRM score of patients is too high (more distributed at the major peak), CDK4/6 inhibitors may not be able to bring additional benefits than endocrine therapy, although the efficacy of treatment was better than that of patients distributed at the minor peak. In general, when using CRM to predict the efficacy of CDK4/6i combined with endocrine therapy as adjuvant therapy for early breast cancer, it is easier to demonstrate additional efficacy by recruiting patients with the CRM scores in the minor peak range.

### Simulation of trial in HR+/HER2+ breast cancer and the potential of the CRM as a companion diagnostic

Twenty-eight HR+/HER2-breast cancer patients were selected for the No.5 trial simulation after meeting the criteria outlined in Table 2c. The No.5 trial aimed to evaluate the efficacy of CDK4/6i (Abemaciclib) plus trastuzumab with or without fulvestrant and standard-of-care chemotherapy of physician’s choice plus trastuzumab in women with advanced breast cancer. The interim report of the trial indicated that the primary endpoint had been reached, with a median follow-up time of 19 months59. The results showed that the combination of Abemaciclib + fulvestrant + trastuzumab significantly improved PFS and ORR of patients when compared with standard chemotherapy + trastuzumab. Specifically, the ORR of Abemaciclib + fulvestrant + trastuzumab group was 36%, while the standard chemotherapy + trastuzumab group only reached 16%59. In the simulation, the CRM scores of the enrolled patients were distributed as a three-peak pattern surrounding 0.87, 0.70, and 0.33, respectively (Figure 4D), indicating that the enrolled patients were heterogeneous in response to CDK4/6i-based therapies. The median score of simulated patients was 0.8181, and patients with scores larger than 0.8 (the dividing point between the higher peaks) accounted for 47% of simulated patients, falling within the 95% confidence interval of the ORR observed in the CDK4/6i + fulvestrant + trastuzumab group in the clinical study. These results suggest that the CRM has potential for establishing companion diagnostics for CDK4/6i-based therapies by screening patients in a proper score range based on their disease and medication status.

## Discussions

In this study, we developed a novel AI approach using the Damage Assessment Framework of Genomic Mutations (DAGM) algorithm to derive pathway-level quantitative information, named as activity profiles of signaling pathways (APSP), from tumor genomes. The information was used for identifying differential features between HR+/HER2- and HR-/HER2-type of breast cancers and predict response to CDK4/6i-based therapies. We trained our model, CRM, on TCGA data and successfully validated it on 343 patients from Guangdong, China. The CRM scores are strongly linked to CDK4/6i responses verified by CDK4/6i-treated patients and five patient clusters were identified by the model with differential responses. These findings also hypothesize that a large proportion of patients with HR+/HER2+ breast cancer might be benefited from CDK4/6i-based therapies. Furthermore, real-world clinical trials simulation showed CRM’s ability to manifest differences in patient responses observed in clinical practice.

Breast cancer patients’ responses to CDK4/6i-based therapies vary widely, making it difficult to predict which patients will benefit from this kind of treatment. To tackle with this complex question, we proposed a novel AI-based method to link genomic data from patients and the responses of CDK4/6i-based therapies. By transforming genomic data to functional information, not only can a lower requirements of computing power be offered for the further machine learning process, but also the underlying mechanisms of responding to CDK4/6i can be investigated.

Moreover, the transformed functional information (APSP) of patients can be considered as a form of digital twins of the real patients. Therefore, our ICT based on personal APSP and the CRM as a digital drug is an accurate simulation of “patients on medication” and can be easily adapted and applied in real clinical practice. As shown in ICT, simulated patients with stage IV NSCLC presented low CRM scores, corresponding with the clinical outcomes. On the basis of our ICT results, the CRM was applied to patients with other cancer and guided the CDK4/6i-treatment in investigator-initiated research, and the current results are proven to be positive (unpublished). These findings encouraged the CRM as a potential companion diagnostic to be applied in pan-cancer treatment.

From the CRM scoring, patients with some diseases might be mostly concentrated at a high range like Luminal A breast cancer. When considering medication scheme, patients with high-score disease exhibit higher possibility for responding to CDK4/6i-based therapy and may not be required to take companion diagnostic test by CRM. For disease with heterogenetic CRM score distributions like HR+/HER2+ patients, the CRM as companion diagnostic is necessary and proper medication scheme can be assigned to suitable patients, which approach the goal of precision medication.

Although we’ve tried to acquire higher quality data, more organized and intact information of disease status and medication records are needed in the future to optimize the CRM. To face with more complicated and refined clinical scenes, detailed pharmacodynamic indicators, like PFS and OS, need to be collected, so that factors truly affected these indicators can be screened and the CRM can be optimized by these factors to benefit patients better.

Additionally, the CRM can be applied to screen suitable patients in different clinical stages or lines of treatment with modifications. For instance, patients with relapsed HR+/HER2-advanced breast cancer after multiple lines of therapies may possess tumors with severe malignancy, and their response rates to various treatments may be significantly lower than patients in other conditions. Thus, adjustments in the CRM score distribution pattern and screening methods may be necessary. In the MONARCH 1 ([NCT02102490](http://medrxiv.org/lookup/external-ref?link_type=CLINTRIALGOV&access_num=NCT02102490&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom)) clinical trial, for evaluating CDK4/6i efficacy in the heavily treated HR+/HER2-patients, the ORR for CDK4/6i was 19.7%60, which is lower than the outcomes of clinical trials for patients in other conditions. This result supported the hypothesis that the CRM score distribution of these patients may be concentrated at the lower range, and the threshold for screening patients in this condition might also be lower than the other conditions.

Other improvements, like enriching the cellular functions inside personal APSP, could be helpful for a more comprehensive discovery of the cellular mechanisms in responding to CDK4/6i based therapies, and therefore improving the accuracy and reliability of the CRM in guiding CDK4/6i-related treatment.

Massive attempts were carried out to expand the usage of CDK4/6i-based therapies61, and the CRM might provide a new prospective to this field. The CRM-like methods carry the ability to investigate the relationship between CDK4/6i-responding mechanism and the cellular mechanism facilitating tumor growth or immune response. By collecting responses to other therapies, like PARP inhibitors and anti-angiogenic therapies, we could establish models like the CRM and find if these therapies having complementary or opposing effects in anti-tumor mechanisms. Hence, the drugs could be rationally assigned to patients and boost the curative capacity of the drugs.

## Methods

### Patient Cohort and Data Collection

In this study, we analyzed the Whole Exome Sequencing (WES) data of 980 breast cancer patients from The Cancer Genome Atlas (TCGA) and 541 breast cancer patients from Guangdong, China. The TCGA cohort was categorized into four subtypes based on the expression of ER (estrogen receptor), PR (progesterone receptor), and HER2, including HR+/HER2-type (Luminal A and Luminal B1 type, n=454), HR+/HER2+ type (Luminal B2 type, n=58), HR-/HER2+ type (HER2-enriched type, n=33), and HR-/HER2-type (TN type, n=127). The Guangdong cohort was also divided into the same four subtypes, including HR+/HER2-(n=206), HR+/HER2+ (n=92), HR-/HER2+ (n=82) and HR-/HER2-(n=137). The classification of breast cancer was based on Goldhirsch’s research on 201362. WES data was obtained from peripheral blood and lesions of the patients to facilitate a comprehensive analysis of somatic mutations in breast cancer genomes. The detailed genomic characteristics of each subtype were explored to identify potential determinants of CDK4/6i-based therapies sensitivity in breast cancer patients.

### CRM establishment, verification and optimization

To establish a prediction model for CDK4/6 inhibitor-based therapy, the baseline characteristics of four patient groups in TCGA were compared. The adapted DAGM algorithm was used to analyze the clonal somatic activity profiles of signaling pathways (sAPSP) based on the genomic data of patients’ tumor tissues and peripheral blood samples. The sAPSP were represented as a list of quantitative measurements of signaling pathways, indicating their activation and inhibition status. The APSP characters were defined as differences in the mean of the APSPs between subtypes, with the Z score used to assess significant differences. APSP characters with an absolute Z score ≥3 were considered as differential features between HR+/HER2- and HR-/HER2-subtypes, reflecting their differences in response mechanism to CDK4/6i-based therapies.

Based on the differential features selected above as model parameters, a prediction model, CRM, was established using the random forest method. The genomic data of HR+/HER2- and HR-/HER2-breast cancer patients in TCGA were used as the training set. This CRM can distinguish the two type of patients by leave-one-out cross validation (LOOCV)63. The closer the score is to HR+/HER2-, the more suitable the patient is for CDK4/6 inhibitors.

To validate the CRM, the local WES data of HR+/HER2- and HR-/HER2-subtypes were used as testing sets. After verification, the CRM’s performance was further improved. The TCGA and Guangdong local database were combined to expand the amount of training data, leading to an enhanced model in distinguishing patients. Moreover, the optimized CRM helped establish a unified set of evaluation methods for the suitability of using CDK4/6 inhibitors for patients with different ancestries.

### Validation of the CRM score and CDK4/6i-efficacy

To validate the linkage between the CRM scores and drug efficacy, 13 female patients with pathologically diagnosed breast cancer were recruited from the Guangdong local database. The patients, aged between 26 and 72 years old, were tested for molecular typing by immunohistochemistry. The 13 patients were all diagnosed with HR+/HER2-type breast cancer. All patients received CDK4/6i (Palbociclib)-based combination therapy, which included one or more drugs in fulvestrant, anastrozole, zoledronic acid, letrozole, exemestane, everolimus, and Norad. These patients were at different stages of treatment, including first-line, second-line, or multi-line post-treatment, and the shortest PFS (progression-free survival) among them was 2 months.

The Response Evaluation Criteria in Solid Tumors (RECIST) was used to evaluate each patient’s response to CDK4/6i-based treatment, and the patients were classified as complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD)64. Objective response rate was used to define patients with CR and PR as responders of CDK4/6i, whereas patients with SD and PD were defined as non-responders of CDK4/6i. The CRM was applied to evaluate these responders and non-responders to check if the CRM could accurately reflect the response differences.

### The CRM prediction on other breast cancer subtypes

To further explore the predictive ability of the CRM, breast cancer patients from TCGA and local databases were scored based on their cancer subtype classification. Specifically, HR+/HER2-patients were further divided into Luminal A type (Ki-67 < 15%) and Luminal B1 type (Ki-67 >= 15%) based on the Ki-67 index measured in local clinical detection and chosen from the study of PAM50 and Claudin-low (CLOW) molecular subtypes for TCGA patients65. The CRM scores were calculated for each subtype, and their distribution characteristics were analyzed to predict the response to CDK4/6i-based therapies in different breast cancer subtypes.

### In-silico clinical trials

#### Clinical Trials Selection

Five clinical trials were chosen for this study, which employed CDK4/6 inhibitors-based therapy to treat breast cancer and non-small cell lung cancer. The No.1 to No.4 trials were selected by distinct outcomes observed clinically, and the No.5 trial was chosen to demonstrate the possible correlation between the CRM scores and drug efficacy. The basic information of the five selected clinical trials is presented in Table 4.

View this table:
[Table 4](http://medrxiv.org/content/early/2023/05/16/2023.05.15.23289976/T4)

Table 4 Basic information of selected clinical trials 9,54,56,58,59

#### Simulated patients screening

To simulate patients enrolled in clinical trials, 950 and 1051 patients with complete clinical information from the TCGA breast cancer and lung cancer database (TCGA-BRCA, TCGA-LUAD, TCGA-LUSC) were included. The specific clinical information of eligible patients in TCGA was selected to screen out digital twins that closely resemble clinical trial patients.

#### Results comparison

The CRM scores of simulated patients in each trial were calculated and plotted to illustrate their distributions. These simulated results were compared across trials and with real-world clinical observations. Additionally, for the No.5 clinical trial, the scores of simulated patients were preliminarily screened to determine if they can accurately reflect the clinical outcomes of patients. This demonstrated the potential use of the CRM as a companion diagnostic to improve precision medicine.

#### About LLM

When constructing the manuscript, we considered the results given by LLM algorithms for the text in some paragraphs, but we did not use them directly.

## Data Availability

All data produced in the present study are available upon reasonable request to the authors

## Author contributions

G. Niu and Y. Liu conceived the study. G. Niu, M. Yang, Y. Liu and Q. Zhang designed the study. M. Yang collected clinical data from patients and provided clinical guidance. G. Niu, Y. Liu, Q. Zhang, Y. Fan, J. Xu and X. Li performed data analysis. Y. Liu, M. Yang, Y. Hsueh and G. Niu wrote the manuscript. All authors reviewed and approved the final version of the manuscript.

## Competing interests

The authors have no competing interests to declare. We have no financial or other relationship with any pharmaceutical companies that produce drugs mentioned in this study, including but not limited to CDK4/6 inhibitors.

## Materials & Correspondence

Correspondence and requests for materials should be addressed to Dr. Gang Niu via g.niu{at}philrivers.com

## Acknowledgements

We thank all the patients who participated in this study and the medical staff who collected clinical data and samples. We also thank Guangdong Provincial People’s Hospital that provided data and clinical guidance for this study. We acknowledge the invaluable contributions of the Philrivers computational biology team that provided critical analysis and interpretation of the data. We thank West Institute of Computing Technology for offering high performance computing platform. We express our gratitude to all the colleagues who contributed to this study through their expertise, insights, and support. This work was supported by National Natural Science Foundation of China (grant no. 82072939 to M. Yang), Natural Science Foundation of Guangdong Province (grant no. 2022A1515010425 to M. Yang), Guangzhou Science and Technology Program (grant no. 202206010110 to M. Yang), Hefei National Laboratory for Physical Sciences at the Microscale (grant no. KF2020009 to G. Niu).

*   Received May 15, 2023.
*   Revision received May 15, 2023.
*   Accepted May 16, 2023.


*   © 2023, Posted by Cold Spring Harbor Laboratory

The copyright holder for this pre-print is the author. All rights reserved. The material may not be redistributed, re-used or adapted without the author's permission.

## References

1.  Braal, C. L. et al. Inhibiting CDK4/6 in Breast Cancer with Palbociclib, Ribociclib, and Abemaciclib: Similarities and Differences. Drugs 81, 317–331 (2021). [https://doi.org:10.1007/s40265-020-01461-2](https://doi.org:10.1007/s40265-020-01461-2)
    
    

2.  Finn, R. S. et al. Palbociclib and Letrozole in Advanced Breast Cancer. N Engl J Med 375, 1925–1936 (2016). [https://doi.org:10.1056/NEJMoa1607303](https://doi.org:10.1056/NEJMoa1607303)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1056/NEJMoa1607303&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=27959613&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

3.  Cristofanilli, M. et al. Fulvestrant plus palbociclib versus fulvestrant plus placebo for treatment of hormone-receptor-positive, HER2-negative metastatic breast cancer that progressed on previous endocrine therapy (PALOMA-3): final analysis of the multicentre, double-blind, phase 3 randomised controlled trial. Lancet Oncol 17, 425–439 (2016). [https://doi.org:10.1016/S1470-2045(15)00613-0](https://doi.org:10.1016/S1470-2045(15)00613-0)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/S1470-2045(15)00613-0&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=26947331&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

4.  Hortobagyi, G. N. et al. Updated results from MONALEESA-2, a phase III trial of first-line ribociclib plus letrozole versus placebo plus letrozole in hormone receptor-positive, HER2-negative advanced breast cancer. Ann Oncol 29, 1541–1547 (2018). [https://doi.org:10.1093/annonc/mdy155](https://doi.org:10.1093/annonc/mdy155)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1093/annonc/mdy155&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

5.  Tripathy, D. et al. Ribociclib plus endocrine therapy for premenopausal women with hormone-receptor-positive, advanced breast cancer (MONALEESA-7): a randomised phase 3 trial. Lancet Oncol 19, 904–915 (2018). [https://doi.org:10.1016/S1470-2045(18)30292-4](https://doi.org:10.1016/S1470-2045(18)30292-4)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/S1470-2045(18)30292-4&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

6.  Slamon, D. J. et al. Ribociclib plus fulvestrant for postmenopausal women with hormone receptor-positive, human epidermal growth factor receptor 2-negative advanced breast cancer in the phase III randomized MONALEESA-3 trial: updated overall survival. Ann Oncol 32, 1015–1024 (2021). [https://doi.org:10.1016/j.annonc.2021.05.353](https://doi.org:10.1016/j.annonc.2021.05.353)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.annonc.2021.05.353&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=34102253&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

7.  De Laurentiis, M. et al. Full population results from the core phase of CompLEEment-1, a phase 3b study of ribociclib plus letrozole as first-line therapy for advanced breast cancer in an expanded population. Breast Cancer Res Treat 189, 689–699 (2021). [https://doi.org:10.1007/s10549-021-06334-0](https://doi.org:10.1007/s10549-021-06334-0)
    
    

8.  Martin, M. et al. Treatment With Adjuvant Abemaciclib Plus Endocrine Therapy in Patients With High-risk Early Breast Cancer Who Received Neoadjuvant Chemotherapy: A Prespecified Analysis of the monarchE Randomized Clinical Trial. JAMA Oncol 8, 1190–1194 (2022). [https://doi.org:10.1001/jamaoncol.2022.1488](https://doi.org:10.1001/jamaoncol.2022.1488)
    
    

9.  Goetz, M. P. et al. MONARCH 3: Abemaciclib As Initial Therapy for Advanced Breast Cancer. J Clin Oncol 35, 3638–3646 (2017). [https://doi.org:10.1200/JCO.2017.75.6155](https://doi.org:10.1200/JCO.2017.75.6155)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1200/JCO.2017.75.6155&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

10. Sledge, G. W. et al. MONARCH 2: Abemaciclib in Combination With Fulvestrant in Women With HR+/HER2-Advanced Breast Cancer Who Had Progressed While Receiving Endocrine Therapy. J Clin Oncol 35, 2875–2884 (2017). [https://doi.org:10.1200/JCO.2017.73.7585](https://doi.org:10.1200/JCO.2017.73.7585)
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

11. Tan, A. R. et al. Trilaciclib plus chemotherapy versus chemotherapy alone in patients with metastatic triple-negative breast cancer: a multicentre, randomised, open-label, phase 2 trial. Lancet Oncol 20, 1587–1601 (2019). [https://doi.org:10.1016/S1470-2045(19)30616-3](https://doi.org:10.1016/S1470-2045(19)30616-3)
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

12. Eroles, P., Bosch, A., Pérez-Fidalgo, J. A. & Lluch, A. Molecular biology in breast cancer: intrinsic subtypes and signaling pathways. Cancer Treat Rev 38, 698–707 (2012). [https://doi.org:10.1016/j.ctrv.2011.11.005](https://doi.org:10.1016/j.ctrv.2011.11.005)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.ctrv.2011.11.005&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=22178455&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

13. Bychkov, D. et al. Deep learning identifies morphological features in breast cancer predictive of cancer ERBB2 status and trastuzumab treatment efficacy. Sci Rep 11, 4037 (2021). [https://doi.org:10.1038/s41598-021-83102-6](https://doi.org:10.1038/s41598-021-83102-6)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1038/s41598-021-83102-6&link_type=DOI) 

14. Pandey, K. et al. Molecular mechanisms of resistance to CDK4/6 inhibitors in breast cancer: A review. Int J Cancer 145, 1179–1188 (2019). [https://doi.org:10.1002/ijc.32020](https://doi.org:10.1002/ijc.32020)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1002/ijc.32020&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=30478914&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

15. Finn, R. S. et al. The cyclin-dependent kinase 4/6 inhibitor palbociclib in combination with letrozole versus letrozole alone as first-line treatment of oestrogen receptor-positive, HER2-negative, advanced breast cancer (PALOMA-1/TRIO-18): a randomised phase 2 study. Lancet Oncol 16, 25–35 (2015). [https://doi.org:10.1016/S1470-2045(14)71159-3](https://doi.org:10.1016/S1470-2045(14)71159-3)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/S1470-2045(14)71159-3&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=25524798&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

16. Turner, N. C. et al. Cyclin E1 Expression and Palbociclib Efficacy in Previously Treated Hormone Receptor-Positive Metastatic Breast Cancer. J Clin Oncol 37, 1169–1178 (2019). [https://doi.org:10.1200/JCO.18.00925](https://doi.org:10.1200/JCO.18.00925)
    
    

17. Herrera-Abreu, M. T. et al. Early Adaptation and Acquired Resistance to CDK4/6 Inhibition in Estrogen Receptor-Positive Breast Cancer. Cancer Res 76, 2301–2313 (2016). [https://doi.org:10.1158/0008-5472.CAN-15-0728](https://doi.org:10.1158/0008-5472.CAN-15-0728)
    
    [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NjoiY2FucmVzIjtzOjU6InJlc2lkIjtzOjk6Ijc2LzgvMjMwMSI7czo0OiJhdG9tIjtzOjUwOiIvbWVkcnhpdi9lYXJseS8yMDIzLzA1LzE2LzIwMjMuMDUuMTUuMjMyODk5NzYuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9) 

18. Berger, A. C. et al. A Comprehensive Pan-Cancer Molecular Study of Gynecologic and Breast Cancers. Cancer Cell 33, 690–705.e699 (2018). [https://doi.org:10.1016/j.ccell.2018.03.014](https://doi.org:10.1016/j.ccell.2018.03.014)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.ccell.2018.03.014&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

19. Mitri, Z. et al. A phase 1 study with dose expansion of the CDK inhibitor dinaciclib (SCH 727965) in combination with epirubicin in patients with metastatic triple negative breast cancer. Invest New Drugs 33, 890–894 (2015). [https://doi.org:10.1007/s10637-015-0244-4](https://doi.org:10.1007/s10637-015-0244-4)
    
    

20. Kashyap, D. et al. Global Increase in Breast Cancer Incidence: Risk Factors and Preventive Measures. Biomed Res Int 2022, 9605439 (2022). [https://doi.org:10.1155/2022/9605439](https://doi.org:10.1155/2022/9605439)
    
    

21. Ibrahim, A. et al. Artificial intelligence in digital breast pathology: Techniques and applications. Breast 49, 267–273 (2020). [https://doi.org:10.1016/j.breast.2019.12.007](https://doi.org:10.1016/j.breast.2019.12.007)
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

22. Piezzo, M. et al. Targeting Cell Cycle in Breast Cancer: CDK4/6 Inhibitors. Int J Mol Sci 21 (2020). [https://doi.org:10.3390/ijms21186479](https://doi.org:10.3390/ijms21186479)
    
    

23. Hamilton, E. & Infante, J. R. Targeting CDK4/6 in patients with cancer. Cancer Treat Rev 45, 129–138 (2016). [https://doi.org:10.1016/j.ctrv.2016.03.002](https://doi.org:10.1016/j.ctrv.2016.03.002)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.ctrv.2016.03.002&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=27017286&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

24. Sherr, C. J., Beach, D. & Shapiro, G. I. Targeting CDK4 and CDK6: From Discovery to Therapy. Cancer Discov 6, 353–367 (2016). [https://doi.org:10.1158/2159-8290.CD-15-0894](https://doi.org:10.1158/2159-8290.CD-15-0894)
    
    [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NzoiY2FuZGlzYyI7czo1OiJyZXNpZCI7czo3OiI2LzQvMzUzIjtzOjQ6ImF0b20iO3M6NTA6Ii9tZWRyeGl2L2Vhcmx5LzIwMjMvMDUvMTYvMjAyMy4wNS4xNS4yMzI4OTk3Ni5hdG9tIjt9czo4OiJmcmFnbWVudCI7czowOiIiO30=) 

25. Spring, L. M. et al. Cyclin-dependent kinase 4 and 6 inhibitors for hormone receptor-positive breast cancer: past, present, and future. Lancet 395, 817–827 (2020). [https://doi.org:10.1016/S0140-6736(20)30165-3](https://doi.org:10.1016/S0140-6736(20)30165-3)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/S0140-6736(20)30165-3&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

26. Dhillon, S. Trilaciclib: First Approval. Drugs 81, 867–874 (2021). [https://doi.org:10.1007/s40265-021-01508-y](https://doi.org:10.1007/s40265-021-01508-y)
    
    

27. Salvador-Barbero, B. et al. CDK4/6 Inhibitors Impair Recovery from Cytotoxic Chemotherapy in Pancreatic Adenocarcinoma. Cancer Cell 37, 340–353.e346 (2020). [https://doi.org:10.1016/j.ccell.2020.01.007](https://doi.org:10.1016/j.ccell.2020.01.007)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.ccell.2020.01.007&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=32109375&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

28. Yang, Y. et al. CDK4/6 inhibitors: a novel strategy for tumor radiosensitization. J Exp Clin Cancer Res 39, 188 (2020). [https://doi.org:10.1186/s13046-020-01693-w](https://doi.org:10.1186/s13046-020-01693-w)
    
    

29. Fassl, A., Geng, Y. & Sicinski, P. CDK4 and CDK6 kinases: From basic science to cancer therapy. Science 375, eabc1495 (2022). [https://doi.org:10.1126/science.abc1495](https://doi.org:10.1126/science.abc1495)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1126/science.abc1495&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=35025636&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

30. Griffiths, J. I. et al. Serial single-cell genomics reveals convergent subclonal evolution of resistance as early-stage breast cancer patients progress on endocrine plus CDK4/6 therapy. Nat Cancer 2, 658–671 (2021). [https://doi.org:10.1038/s43018-021-00215-7](https://doi.org:10.1038/s43018-021-00215-7)
    
    

31. Al-Khafaji, K. & Taskin Tok, T. Amygdalin as multi-target anticancer drug against targets of cell division cycle: double docking and molecular dynamics simulation. J Biomol Struct Dyn 39, 1965–1974 (2021). [https://doi.org:10.1080/07391102.2020.1742792](https://doi.org:10.1080/07391102.2020.1742792)
    
    

32. Zhao, M. et al. Combining Neratinib with CDK4/6, mTOR, and MEK Inhibitors in Models of HER2-positive Cancer. Clin Cancer Res 27, 1681–1694 (2021). [https://doi.org:10.1158/1078-0432.CCR-20-3017](https://doi.org:10.1158/1078-0432.CCR-20-3017)
    
    [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6MTA6ImNsaW5jYW5yZXMiO3M6NToicmVzaWQiO3M6OToiMjcvNi8xNjgxIjtzOjQ6ImF0b20iO3M6NTA6Ii9tZWRyeGl2L2Vhcmx5LzIwMjMvMDUvMTYvMjAyMy4wNS4xNS4yMzI4OTk3Ni5hdG9tIjt9czo4OiJmcmFnbWVudCI7czowOiIiO30=) 

33. Freeman-Cook, K. et al. Expanding control of the tumor cell cycle with a CDK2/4/6 inhibitor. Cancer Cell 39, 1404–1421.e1411 (2021). [https://doi.org:10.1016/j.ccell.2021.08.009](https://doi.org:10.1016/j.ccell.2021.08.009)
    
    

34. Slamon, D. J. et al. Phase III Randomized Study of Ribociclib and Fulvestrant in Hormone Receptor-Positive, Human Epidermal Growth Factor Receptor 2-Negative Advanced Breast Cancer: MONALEESA-3. J Clin Oncol 36, 2465–2472 (2018). [https://doi.org:10.1200/JCO.2018.78.9909](https://doi.org:10.1200/JCO.2018.78.9909)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1200/JCO.2018.78.9909&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

35. Wu, X. et al. Distinct CDK6 complexes determine tumor cell response to CDK4/6 inhibitors and degraders. Nat Cancer 2, 429–443 (2021). [https://doi.org:10.1038/s43018-021-00174-z](https://doi.org:10.1038/s43018-021-00174-z)
    
    

36. Bertucci, F. et al. Genomic characterization of metastatic breast cancers. Nature 569, 560–564 (2019). [https://doi.org:10.1038/s41586-019-1056-z](https://doi.org:10.1038/s41586-019-1056-z)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1038/s41586-019-1056-z&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=31118521&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

37. Li, Z. et al. Loss of the FAT1 Tumor Suppressor Promotes Resistance to CDK4/6 Inhibitors via the Hippo Pathway. Cancer Cell 34, 893–905.e898 (2018). [https://doi.org:10.1016/j.ccell.2018.11.006](https://doi.org:10.1016/j.ccell.2018.11.006)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.ccell.2018.11.006&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

38. Razavi, P. et al. The Genomic Landscape of Endocrine-Resistant Advanced Breast Cancers. Cancer Cell 34, 427–438.e426 (2018). [https://doi.org:10.1016/j.ccell.2018.08.008](https://doi.org:10.1016/j.ccell.2018.08.008)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.ccell.2018.08.008&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=30205045&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

39. Formisano, L. et al. Aberrant FGFR signaling mediates resistance to CDK4/6 inhibitors in ER+ breast cancer. Nat Commun 10, 1373 (2019). [https://doi.org:10.1038/s41467-019-09068-2](https://doi.org:10.1038/s41467-019-09068-2)
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

40. Álvarez-Fernández, M. & Malumbres, M. Mechanisms of Sensitivity and Resistance to CDK4/6 Inhibition. Cancer Cell 37, 514–529 (2020). [https://doi.org:10.1016/j.ccell.2020.03.010](https://doi.org:10.1016/j.ccell.2020.03.010)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.ccell.2020.03.010&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

41. Migliaccio, I. et al. CDK4/6 inhibitors: A focus on biomarkers of response and post-treatment therapeutic strategies in hormone receptor-positive HER2-negative breast cancer. Cancer Treat Rev 93, 102136 (2021). [https://doi.org:10.1016/j.ctrv.2020.102136](https://doi.org:10.1016/j.ctrv.2020.102136)
    
    

42. Vorwerk, J. et al. Presence of the. Front Oncol 12, 903691 (2022). [https://doi.org:10.3389/fonc.2022.903691](https://doi.org:10.3389/fonc.2022.903691)
    
    

43. Furrer, D. et al. Evaluation of human epidermal growth factor receptor 2 (HER2) single nucleotide polymorphisms (SNPs) in normal and breast tumor tissues and their link with breast cancer prognostic factors. Breast 30, 191–196 (2016). [https://doi.org:10.1016/j.breast.2016.09.014](https://doi.org:10.1016/j.breast.2016.09.014)
    
    

44. He, Y., Liu, H., Chen, Q., Shao, Y. & Luo, S. Relationships between SNPs and prognosis of breast cancer and pathogenic mechanism. Mol Genet Genomic Med 7, e871 (2019). [https://doi.org:10.1002/mgg3.871](https://doi.org:10.1002/mgg3.871)
    
    

45. O’Leary, B., Finn, R. S. & Turner, N. C. Treating cancer with selective CDK4/6 inhibitors. Nat Rev Clin Oncol 13, 417–430 (2016). [https://doi.org:10.1038/nrclinonc.2016.26](https://doi.org:10.1038/nrclinonc.2016.26)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1038/nrclinonc.2016.26&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=27030077&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

46. Goel, S., DeCristo, M. J., McAllister, S. S. & Zhao, J. J. CDK4/6 Inhibition in Cancer: Beyond Cell Cycle Arrest. Trends Cell Biol 28, 911–925 (2018). [https://doi.org:10.1016/j.tcb.2018.07.002](https://doi.org:10.1016/j.tcb.2018.07.002)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.tcb.2018.07.002&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=30061045&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

47. Goel, S. et al. CDK4/6 inhibition triggers anti-tumour immunity. Nature 548, 471–475 (2017). [https://doi.org:10.1038/nature23465](https://doi.org:10.1038/nature23465)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1038/nature23465&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=28813415&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

48. Teh, J. L. F. & Aplin, A. E. Arrested Developments: CDK4/6 Inhibitor Resistance and Alterations in the Tumor Immune Microenvironment. Clin Cancer Res 25, 921–927 (2019). [https://doi.org:10.1158/1078-0432.CCR-18-1967](https://doi.org:10.1158/1078-0432.CCR-18-1967)
    
    [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6MTA6ImNsaW5jYW5yZXMiO3M6NToicmVzaWQiO3M6ODoiMjUvMy85MjEiO3M6NDoiYXRvbSI7czo1MDoiL21lZHJ4aXYvZWFybHkvMjAyMy8wNS8xNi8yMDIzLjA1LjE1LjIzMjg5OTc2LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ==) 

49. Yang, M. et al. DAGM: A novel modelling framework to assess the risk of HER2-negative breast cancer based on germline rare coding mutations. EBioMedicine 69, 103446 (2021). [https://doi.org:10.1016/j.ebiom.2021.103446](https://doi.org:10.1016/j.ebiom.2021.103446)
    
    

50. Hu, J. & Szymczak, S. A review on longitudinal data analysis with random forest. Brief Bioinform 24 (2023). [https://doi.org:10.1093/bib/bbad002](https://doi.org:10.1093/bib/bbad002)
    
    

51. Gutiérrez-Casares, J. R. et al. Methods to Develop an. Front Psychiatry 12, 741170 (2021). [https://doi.org:10.3389/fpsyt.2021.741170](https://doi.org:10.3389/fpsyt.2021.741170)
    
    

52. Sips, F. L. P., Pappalardo, F., Russo, G. & Bursi, R. In silico clinical trials for relapsing-remitting multiple sclerosis with MS TreatSim. BMC Med Inform Decis Mak 22, 294 (2022). [https://doi.org:10.1186/s12911-022-02034-x](https://doi.org:10.1186/s12911-022-02034-x)
    
    

53. Johnston, S. et al. MONARCH 3 final PFS: a randomized study of abemaciclib as initial therapy for advanced breast cancer. NPJ Breast Cancer 5, 5 (2019). [https://doi.org:10.1038/s41523-018-0097-z](https://doi.org:10.1038/s41523-018-0097-z)
    
    

54. Kim, E. S. et al. Abemaciclib in Combination with Single-Agent Options in Patients with Stage IV Non-Small Cell Lung Cancer: A Phase Ib Study. Clin Cancer Res 24, 5543–5551 (2018). [https://doi.org:10.1158/1078-0432.CCR-18-0651](https://doi.org:10.1158/1078-0432.CCR-18-0651)
    
    [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6MTA6ImNsaW5jYW5yZXMiO3M6NToicmVzaWQiO3M6MTA6IjI0LzIyLzU1NDMiO3M6NDoiYXRvbSI7czo1MDoiL21lZHJ4aXYvZWFybHkvMjAyMy8wNS8xNi8yMDIzLjA1LjE1LjIzMjg5OTc2LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ==) 

55. Toi, M. et al. Adjuvant Abemaciclib Combined with Endocrine Therapy: Efficacy Results in monarchE Cohort 1. Oncologist 28, e77–e81 (2023). [https://doi.org:10.1093/oncolo/oyac234](https://doi.org:10.1093/oncolo/oyac234)
    
    

56. Gnant, M. et al. Adjuvant Palbociclib for Early Breast Cancer: The PALLAS Trial Results (ABCSG-42/AFT-05/BIG-14-03). J Clin Oncol 40, 282–293 (2022). [https://doi.org:10.1200/JCO.21.02554](https://doi.org:10.1200/JCO.21.02554)
    
    

57. Harbeck, N. et al. Adjuvant abemaciclib combined with endocrine therapy for high-risk early breast cancer: updated efficacy and Ki-67 analysis from the monarchE study. Ann Oncol 32, 1571–1581 (2021). [https://doi.org:10.1016/j.annonc.2021.09.015](https://doi.org:10.1016/j.annonc.2021.09.015)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.annonc.2021.09.015&link_type=DOI) 

58. Johnston, S. R. D. et al. Abemaciclib Combined With Endocrine Therapy for the Adjuvant Treatment of HR+, HER2-, Node-Positive, High-Risk, Early Breast Cancer (monarchE). J Clin Oncol 38, 3987–3998 (2020). [https://doi.org:10.1200/JCO.20.02514](https://doi.org:10.1200/JCO.20.02514)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1200/jco.20.02514&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=32954927&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

59. Tolaney, S. M. et al. Abemaciclib plus trastuzumab with or without fulvestrant versus trastuzumab plus standard-of-care chemotherapy in women with hormone receptor-positive, HER2-positive advanced breast cancer (monarcHER): a randomised, open-label, phase 2 trial. Lancet Oncol 21, 763–775 (2020). [https://doi.org:10.1016/S1470-2045(20)30112-1](https://doi.org:10.1016/S1470-2045(20)30112-1)
    
    

60. Rugo, H. S. et al. Real-world survival outcomes of heavily pretreated patients with refractory HR+, HER2-metastatic breast cancer receiving single-agent chemotherapy-a comparison with MONARCH 1. Breast Cancer Res Treat 184, 161–172 (2020). [https://doi.org:10.1007/s10549-020-05838-5](https://doi.org:10.1007/s10549-020-05838-5)
    
    

61. Lim, J. S., Turner, N. C. & Yap, T. A. CDK4/6 Inhibitors: Promising Opportunities beyond Breast Cancer. Cancer Discov 6, 697–699 (2016). [https://doi.org:10.1158/2159-8290.CD-16-0563](https://doi.org:10.1158/2159-8290.CD-16-0563)
    
    [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NzoiY2FuZGlzYyI7czo1OiJyZXNpZCI7czo3OiI2LzcvNjk3IjtzOjQ6ImF0b20iO3M6NTA6Ii9tZWRyeGl2L2Vhcmx5LzIwMjMvMDUvMTYvMjAyMy4wNS4xNS4yMzI4OTk3Ni5hdG9tIjt9czo4OiJmcmFnbWVudCI7czowOiIiO30=) 

62. Goldhirsch, A. et al. Personalizing the treatment of women with early breast cancer: highlights of the St Gallen International Expert Consensus on the Primary Therapy of Early Breast Cancer 2013. Ann Oncol 24, 2206–2223 (2013). [https://doi.org:10.1093/annonc/mdt303](https://doi.org:10.1093/annonc/mdt303)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1093/annonc/mdt303&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=23917950&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000323963100003&link_type=ISI) 

63. Poldrack, R. A., Huckins, G. & Varoquaux, G. Establishment of Best Practices for Evidence for Prediction: A Review. JAMA Psychiatry 77, 534–540 (2020). [https://doi.org:10.1001/jamapsychiatry.2019.3671](https://doi.org:10.1001/jamapsychiatry.2019.3671)
    
    

64. Schwartz, L. H. et al. RECIST 1.1-Update and clarification: From the RECIST committee. Eur J Cancer 62, 132–137 (2016). [https://doi.org:10.1016/j.ejca.2016.03.081](https://doi.org:10.1016/j.ejca.2016.03.081)
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.ejca.2016.03.081&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=27189322&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F05%2F16%2F2023.05.15.23289976.atom) 

65. Thennavan, A. et al. Molecular analysis of TCGA breast cancer histologic types. Cell Genom 1 (2021). [https://doi.org:10.1016/j.xgen.2021.100067](https://doi.org:10.1016/j.xgen.2021.100067)