Palbociclib plus endocrine therapy significantly enhances overall survival of HR+/HER2− metastatic breast cancer patients compared to endocrine therapy alone in the second-line setting-a large institutional study

Abstract

Cyclin-dependent-kinase-4/6 inhibitor (CDKi) plus endocrine therapy (ET) is standard of care for patients with advanced hormone receptor (HR)-positive, HER2-negative breast cancer (BC). The Breast Medical Oncology database at MD Anderson Cancer Center (MDACC) was analyzed to assess effectiveness of the CDKi palbociclib plus ET compared to ET alone. From a total of 5402 advanced HR+ HER2− BC patients referred to MDACC between 1997 and 2020, we identified eligible patients who received palbociclib in combination with first- (n=778) and second-line (n=410) ET. We further identified “control” patients who received ET alone in the first- (n=2452) and second-line (n=1183) settings. Propensity score matching analysis was conducted to balance baseline demographic and clinical characteristics between palbociclib and control cohorts to assess the effect of palbociclib treatment on progression-free survival (PFS) and overall survival (OS). For propensity-matched-cohort in the first-line setting (n=708), palbociclib group had significantly longer median PFS (17.4 vs. 11.1 months; p<0.0001) compared to controls. Median OS (44.3 vs. 40.2 months) did not show a statistically significant benefit in the first line setting. However, in the second-line setting, with 380 propensity-matched-cohort, the palbociclib group had significantly longer PFS (10 vs 5 months, p<0.0001) as well as OS (33 vs 24 months; p < 0.022), compared to controls. We conclude that in this single center analysis of a large cohort of metastatic HR+ HER2− BC patients, palbociclib in combination with ET was associated with improved PFS in both first- and second-line settings and OS in the second-line setting compared with ET alone cohort.

Graphical abstract

Introduction

Breast cancer is the most frequently diagnosed malignancy with an estimated 1.7 million new diagnoses/year (23% of all cancer diagnoses) worldwide ¹. After receiving curative surgery with or without systemic therapy, about 20-50% of all breast cancer patients develop recurrence². Despite significant recent clinical progress in outcomes for metastatic breast cancer (MBC), it remains a deadly disease, with a profound negative impact on patients’ quality and length of life³.

Over 70% of newly diagnosed breast cancers are hormone receptor-positive (HR+) (estrogen and/or progesterone receptor) and HER2-negative ⁴ such that a majority of deaths due to breast cancer are from HR+ MBC ⁵. Growth factor pathway activation has been shown to be a driver of resistance to endocrine therapy (ET), and recent advances in biologically targeted therapies against mechanistic targets of rapamycin (mTOR), phosphoinositide 3-kinase (PI3 kinase)^{6, 7}, and cyclin-dependent kinase 4/6 (CDK4/6), have proven successful in delaying progression in patients with MBC when added to either aromatase inhibitors (AIs) or the selective estrogen receptor degrader, fulvestrant (collectively called ET) ^8–15. Three CDK4/6 inhibitors (CDK4/6i), palbociclib, ribociclib, and abemaciclib, are approved in the first- or second-line settings in combination with ET on the basis of increased progression-free survival (PFS) as compared to ET alone ^{8, 11, 13, 14, 16–19}. Improved survival has also been reported in some of these clinical trials ^20–22. The PFS and OS from these prior clinical trials is summarized in Table S1. In PALOMA-1, patients who received palbociclib + letrozole had a median PFS of 20.2 months, compared with only 10.2 months with letrozole alone (hazard ratio [HR], 0.488; 95% CI, 0.319–0.748; P = .0004) ⁸. Median overall survival (OS) was 37.5 months with palbociclib + letrozole and 34.5 months with letrozole alone (HR, 0.897; 95% CI, 0.623–1.294; P = .281). This difference was not statistically significant, but the study was not powered to show a difference in OS ²³. In the phase-3 PALOMA-2 study, median PFS was 24.8 months in patients who received palbociclib + letrozole and 14.5 months in those who received placebo and letrozole (HR, 0.58; 95% CI, 0.46–0.72; P = .001) ¹⁰. OS data for PALOMA-2 is still maturing. In PALOMA-3, the median PFS in the palbociclib + fulvestrant group of 9.5 months was significantly higher than the 4.6 months in the placebo and fulvestrant group (HR, 0.46; 95% CI, 0.36–0.59; P < .0001) ^{11, 24}. Among the 410 patients with documented sensitivity to previous ET, the median OS was 39.7 months with palbociclib + fulvestrant and 29.7 months with fulvestrant alone in an exploratory analysis (HR, 0.72; 95% CI, 0.55 to 0.94) ²⁵. Improvements in OS continued to be observed in the PALOMA-3 patients with longer follow up time of >5 years ²⁶ and median OS of 34.9 months (HR, 0.81; 95% CI, 0.654–0.94l 1-sided nominal p= 0.0221) ²⁷ (Table S1). The estimated 5-year OS rates was 23.3% with palbociclib as compared to 16.8% with placebo²⁶.

Similar results were observed with the ribociclib and abemaciclib clinical trials. A consistent PFS gain from 12.8 months to 20.5 months was reported in MONALEESA-3 with ribociclib and from 9.3 months to 16.9 months in Monarch-2 with abemaciclib ^{11, 28}. Results from the phase-3 MONALEESA-2 trial showed a significant OS benefit with ribociclib plus ET for postmenopausal patients with HR positive, HER2-negative MBC ²⁹. Specifically, an OS advantage of 63.9 months with front-line ribociclib and letrozole, compared to 51.4 months with hormone therapy alone was observed. The estimated 6-year survival rate was 44.2% with ribociclib compared with 32% for placebo (Table S1). There was also an OS benefit in patients with abemaciclib + fulvestrant (46.7 months) compared to placebo + fulvestrant (37.3 months) ²¹.

Following the PALOMA clinical trials, several published real-world data on the safety and effectiveness of the CDK4/6 inhibitors + ET in routine clinical practice have collectively, shown similar efficacy and safety profiles to the clinical trials ^30–36. However, these reports were limited in scope due to the heterogenous follow-up practices observed across different institutions; some studies had small sample sizes or lack of a proper “control” cohort. To overcome these limitations, the goal in this study is to provide evidence from a single institution, the MD Anderson Cancer Center, for both PFS and OS of palbociclib + ET using a comprehensively annotated and curated database, and comparing the outcome data to the historical controls using propensity-matched patients who received ET alone in both first-line and second-line settings.

METHODS

Patient Populations

The Department of Breast Medical Oncology database captures all breast cancer patients seen at the MD Anderson Cancer Center (MDACC) since the 1980s (n>53,000 patients). To identify patients for the current study, the database was queried for the following criteria: age ≥ 18 years, referred to MDACC between January 1997 and January 2020 (Figure S1), estrogen receptor (ER) and/or progesterone receptor (PR)-positive and HER2-negative disease, termed as HR+/HER2-MBC. Patients were identified by the following database variables: disease recurrence, stage IV, and TNM stage M1. First- and second-line MBC treatment with ET (AIs or fulvestrant) alone or in combination with palbociclib were also included. If any of the variables were unknown, the patient was excluded from analysis. Patients treated with chemotherapy as first-line metastatic treatment were also excluded. ET included a single AI as initial treatment (anastrozole, exemestane or letrozole) and second line fulvestrant after AI only. We did not include patients who may have received fulvestrant as first line of ET (either in the CDKi or ET-only groups) due to having had recent (<1–2 years) treatment of adjuvant AI. Patients who received abemaciclib or ribociclib, as the CDK4/6i, were also excluded. The STROBE (Strengthening The Reporting of Observational studies in Epidemiology) diagram in Figure 1 depicts the inclusion and exclusion criteria for the different cohorts used in this study.

[Figure 1].

STROBE Diagram

Demographic, clinical and pathologic data collected included age, race, primary stage (i.e., initial stage at presentation), clinical nodal status, metastasis site(s), neoadjuvant and/or adjuvant treatment history, metastatic treatment history and disease status at the time of last contact or death. Clinical stage designation at the initial diagnosis was determined according to American Joint Committee on Cancer (AJCC) guideline version in use at that time ^37–39. HR+ was defined as either ER+ or PR+ by immunohistochemistry (IHC) using institutional cutoffs in accordance with the contemporaneous American Society of Clinical Oncology / College of American Pathologists (ASCO/CAP) guidelines ⁴⁰. HER2 status was determined using either IHC and/or fluorescence in situ hybridization (FISH) ^{41, 42}. Sites of recurrence were defined as non-visceral (local-regional/distant lymph nodes, bone, or soft tissue) or visceral (lung, liver, and other organs). Treatment in the first- and second-line metastatic settings was recorded for each patient.

Treatments and Definitions

To assess the effect of palbociclib in combination with first- and second-line ET, four treatment groups were defined as follows: In the first-line setting, group 1 and group 2 were patients with newly diagnosed metastatic disease, who were treated with AI alone or palbociclib + AI, respectively. In the second-line setting, group 3 and group 4 were defined as patients who received fulvestrant alone or palbociclib + fulvestrant, respectively, following disease progression on first line AI for metastatic disease. Endpoints included progression-free survival (PFS) and overall survival (OS). PFS and OS for these patients with metastatic disease were defined per the recommendations of the National Cancer Institute Breast Cancer Steering Committee Working Group⁴³. Specifically, PFS was defined as the length of time from start of treatment of AI alone or palbociclib + AI to progression for the first-line therapy (PFS at first-line) or death ⁴³. For the second-line setting, PFS was defined as the length of time from start of treatment of fulvestrant alone or palbociclib + fulvestrant (PFS at second-line) to progression or death ⁴³. Progression in both first- and second-line settings was based on physician judgement and documentation based on RECISIT 1.1 ⁴⁴. Patients who were on treatment or for whom no information on end date of treatment was available, or did not have documentation of progression, were considered to be right censored at the time of last follow-up. OS was defined as the length of time from start of treatment with AI alone or palbociclib + AI to death from any cause or right censored for the first-line therapy (OS at first-line) ⁴³. Similarly, OS for the second-line setting was defined as the length of time from the start of treatment with fulvestrant alone or palbociclib + fulvestrant to death due to any cause; surviving patients were right censored at the time of last follow-up ⁴³.

Statistical Analysis

Propensity Score Matching:

To reduce the impact of treatment selection bias on the estimation of therapy effect of palbociclib on PFS and OS using the observational data, we performed propensity score matched analysis to account for baseline differences between control and palbociclib treated patients in both the first-line and second-line therapy patient cohorts ^45–47. The propensity score was defined by the probability of treatment assignment conditional on the observed baseline covariates. For each of the first- and second-line groups, we fitted a logistic regression model to estimate propensity score including baseline factors such as age at diagnosis, race, clinical tumor category, clinical nodal category, clinical metastasis category, ER status, PR status, lymphovascular invasion, or receipt of (neo) adjuvant chemotherapy. Next, we matched each ET + palbociclib treated patient to a control cohort (1:1 matching) based on the estimated propensity scores. The venn diagrams (Figure S2) represent how 3.635 patients who received first-line AI alone (control first-line) paired with 708 patients out of 778 who received first-line AI + palbociclib (palbociclib first-line). In the second-line, 1,183 patients who received fulvestrant alone following first-line AI (control second-line) paired with 380 patients out of 410 patients who received second-line fulvestrant + palbociclib following first-line AI (palbociclib second-line). Due to incomplete information on the baseline covariates that were used for calibrating propensity scores, palbociclib cohorts were reduced causing exclusion of 70 patients in first-line and 30 patients in second-line. We used absolute standardized difference (ASD) of sample means to assess balance in each covariate between palbociclib untreated and treated groups ⁴⁸. The smaller the ASD value, the better the matching for the measured baseline covariates between the two treatment cohorts with ASD values < 0.1 suggesting a good balance in the baseline covariates.

Survival Analyses:

The distributions of PFS and OS, and the 95% confidence intervals (CIs) were estimated by the Kaplan-Meier method ⁴⁹ and the conventional log-rank test ⁵⁰, was performed to evaluate the differences in PFS and OS between treated and untreated groups without propensity score matching. Stratified log-rank test was used to evaluate the difference in PFS or OS for the matched cohorts using each matched pair as a cluster. Regression analyses of survival data, based on the Cox proportional hazards (PH) model, was conducted on PFS and OS outcomes in the multivariate setting. The double robust estimation under the Cox models were used to assess the treatment effect based on the matched cohorts⁵¹. To investigate instantaneous event risk over time, treatment-specific nonparametric smoothed estimates of hazard function (hazard rate over time) was obtained from the bshazard method ³⁸. In the subgroup analysis, we also used landmark analysis ⁵² to assess if palbociclib has OS benefit in the “long-term responders” in the first- and second-line settings. The long-term responders were defined as those patients whose PFS times are longer than the median PFS in each setting. The landmark analyses was conducted conditioning on patients surviving without progression to the designated median PFS times to adjust for potential selection bias. All tests were two-sided. P-values less than 0.05 were considered statistically significant. All the analyses were performed using R version 4.0.2.

RESULTS

Patient Characteristics

Figure 1 presents the STROBE (Strengthening The Reporting of Observational studies in Epidemiology) flow diagram describing the study population. We initially identified 5,402 patients, based on age, referral date (between January 1997 and January 2020), metastatic status and treatment with either ET alone or in combination with CDK4/6 inhibitors. From this cohort, 528 patients were excluded due to incomplete information on their treatment regimens (n=432), or if they had received multiple first line AI treatments prior to 2^nd line therapy (n=96). From the remaining 4874 patients, we excluded an additional 51 patients who had received either abemaciclib (n=29) or ribociclib (n=22). The remaining 4,823 eligible patients included (i) 2,452 patients who received first-line AI alone (control group-first-line) (ii) 778 patients who received first-line AI + palbociclib (treatment group-first-line); (iii) 1,183 patients who received second-line fulvestrant alone following first-line AI (control group for both first and second-line cohorts) and (iv) 410 patients who received second-line fulvestrant + palbociclib following first-line AI (treatment group, second-line). All fulvestrant treated patients (in the control and treatment groups) initially received a single first-line AI without palbociclib. These eligible patients where then divided into 4,413 for first-line analysis and 1,593 for the second-line analyses. Following propensity score matching, 708 paired first-line control and treatment groups, and 380 paired second-line control and treatment groups were identified (blue shaded boxes).

The median OS follow-up duration for the first- and second-line cohorts before and after propensity matching are presented in Table S2 and are as follows, before propensity matching: in the first-line setting 29.4 months (28-31.1) for the treatment group and 118.9 months (110-122.8) for the control group; in the second-line setting 30.1 months (26.4-32.5) for the treatment group, and 119.8 (108-not achieved) months for the control group. Following propensity matching the median OS follow-up duration were as follows: in the first-line setting 30 months (28-31.7) for the treatment group and 119 months (108-128.8) for the matched control group; in the second-line setting 30.1 months (26.3-32.4) for the treatment group and 105.7 months (79.1-not achieved) for the matched control group.

The absolute standardized differences (ASD) of covariates between ET alone and ET + palbociclib in the first-line and second-line settings before and after propensity score matching is depicted in Figures 2A and 2B, respectively. Demographic and clinical characteristics were very different between the first-line palbociclib + AI and AI alone in the unadjusted cohorts (i.e. before propensity matching) but achieved a good balance in the adjusted cohorts after propensity matching in 708 paired patients (Table 1A and Figure 2A). For example, in the unadjusted cohort, the percent PR positivity was at 77.7 and 89.6 between the AI alone and palbociclib + AI groups, respectively, yielding an ASD score of 0.325. However following propensity matching, the percent PR positivity was near identical at 87.0 and 89.0 between the AI alone and palbociclib plus AI, with a ASD of 0.061. For all other baseline covariates, following propensity matching the ASD scores were all <0.1 for other (many well below 0.05), suggesting very well-matched cohorts. Similarly, the ASD scores of <0.1 of the paired 380 patients in the second-line fulvestrant alone and palbociclib + fulvestrant cohort, indicates well-balanced cohorts after the propensity score adjustment (Table 1B and Figure 2B). Thus, the ability of the propensity score to effectively balance the groups according to input variables reveals that in all cases the variables are well matched with an ASD of less than 0.1.

[Figure 2] –

Absolute standardized differences (ASD, probability) in observed covariates between ET alone and palbociclib + ET in the first-line (A) and second-line (B) settings (based on race, age, estrogen or progesterone receptor, neoadjuvant chemotherapy use, lymphovascular invasion, clinical tumor, nodal and metastasis categories, and adjuvant chemotherapy use); the covariates with postmatch (green dots) absolute standardized differences below 0.1 are considered as successfully balanced.

[Table 1A].

Patient demographic and baseline clinical characteristics with absolute standardized difference (ASD) before and after propensity matching for the first-line treatment analysis (ASD<0.1 indicates well-matched cohorts).

	Unadjusted Cohort			Cohort After Propensity Matching
	Control (AI alone)	Palbociclib + AI	ASD	Control (AI alone)	Palbociclib + AI	ASD
N	3635	778		708	708
Age, mean (SD)	50.33 (12.22)	50.31 (11.88)	0.002	50.65 (11.99)	50.65 (11.98)	<0.001
Race (%)			0.102			0.076
White	2762 (76.0)	571 (73.4)		506 (71.5)	520 (73.4)
Black	367 (10.1)	72 (9.3)		57 (8.1)	63 (8.9)
Hispanic and Latino	333 (9.2)	83 (10.7)		86 (12.1)	76 (10.7)
Other	173 (4.8)	52 (6.7)		59 (8.3)	49 (6.9)
Clinical tumor category* (%)			0.151			0.034
T0,T1	1229 (36.4)	219 (30.8)		226 (31.9)	218 (30.8)
T2	1381 (40.9)	333 (46.8)		321 (45.3)	331 (46.8)
T3	364 (10.8)	88 (12.4)		92 (13.0)	88 (12.4)
T4	406 (12.0)	72 (10.1)		69 (9.7)	71 (10.0)
Clinical nodal category* (%)			0.116			0.038
N0	1163 (34.0)	248 (33.8)		241 (34.0)	232 (32.8)
N1	1524 (44.6)	316 (43.1)		298 (42.1)	307 (43.4)
N2	331 (9.7)	58 (7.9)		61 (8.6)	57 (8.1)
N3	401 (11.7)	112 (15.3)		108 (15.3)	112 (15.8)
Clinical metastasis category* - M1 (%)	1123 (31.2)	233 (31.4)	0.004	220 (31.1)	223 (31.5)	0.009
ER Positive (%)	3486 (97.6)	759 (97.6)	<0.001	686 (96.9)	691 (97.6)	0.043
PR Positive (%)	2639 (77.7)	697 (89.6)	0.325	616 (87.0)	630 (89.0)	0.061
Lymphovascular invasion - Yes (%)	1006 (36.8)	237 (30.5)	0.134	247 (34.9)	226 (31.9)	0.063
Adjuvant chemotherapy - Yes (%)	1514 (41.7)	272 (35.0)	0.138	257 (36.3)	254 (35.9)	0.009
Neoadjuvant chemotherapy - Yes (%)	551 (15.2)	173 (22.2)	0.182	163 (23.0)	155 (21.9)	0.027

^*Evaluated at the initial diagnosis

[Table 1B].

Patient demographic and baseline clinical characteristics with absolute standardized difference (ASD) before and after propensity matching for the second-line treatment analysis (ASD<0.1 indicates well-matched cohorts).

	Unadjusted Cohort			Cohort After Propensity Matching
	Control (Fulvestrant alone)	Palbociclib + Fulvestrant	ASD	Control (Fulvestrant alone)	Palbociclib + Fulvestrant	ASD
N	1183	410		380	380
Age, mean (SD)	50.41 (11.74)	51.73 (12.17)	0.11	52.11 (11.77)	51.87 (12.27)	0.02
Race (%)			0.178			0.057
White	923 (78.0)	300 (73.2)		286 (75.3)	283 (74.5)
Black	105 (8.9)	30 (7.3)		29 (7.6)	26 (6.8)
Hispanic and Latino	103 (8.7)	52 (12.7)		38 (10.0)	44 (11.6)
Other	52 (4.4)	28 (6.8)		27 (7.1)	27 (7.1)
Clinical tumor category* (%)			0.172			0.012
T0,T1	419 (38.9)	119 (31.2)		118 (31.1)	119 (31.3)
T2	441 (40.9)	169 (44.2)		168 (44.2)	167 (43.9)
T3	111 (10.3)	51 (13.4)		50 (13.2)	51 (13.4)
T4	106 (9.8)	43 (11.3)		44 (11.6)	43 (11.3)
Clinical nodal category* (%)			0.128			0.077
N0	390 (35.3)	133 (33.9)		118 (31.1)	125 (32.9)
N1	502 (45.4)	170 (43.4)		163 (42.9)	166 (43.7)
N2	102 (9.2)	33 (8.4)		41 (10.8)	33 (8.7)
N3	112 (10.1)	56 (14.3)		58 (15.3)	56 (14.7)
Clinical metastasis category* - M1 (%)	313 (26.7)	73 (18.5)	0.197	73 (19.2)	67 (17.6)	0.041
ER Positive (%)	1141 (98.3)	404 (98.5)	0.021	376 (98.9)	374 (98.4)	0.046
PR -Positive (%)	871 (80.0)	342 (83.4)	0.089	315 (82.9)	317 (83.4)	0.014
Lymphovascular invasion - Yes (%)	328 (35.8)	152 (37.2)	0.028	147 (38.7)	146 (38.4)	0.005
Adjuvant chemotherapy - Yes (%)	530 (44.8)	168 (41.0)	0.077	167 (43.9)	163 (42.9)	0.021
Neoadjuvant chemotherapy - Yes (%)	171 (14.5)	124 (30.2)	0.386	105 (27.6)	117 (30.8)	0.069

^*Evaluated at the initial diagnosis

Progression-free and overall survival benefit

Using the propensity-matched cohorts, we interrogated whether treatment of patients with palbociclib in combination with ET was associated with a longer PFS and OS as compared to the historical control cohorts. In both the first- and second-line settings, patients who received palbociclib had longer PFS (Figure 3). Median PFS was 17.4 months (95% CI 15.9-20) in the first-line palbociclib + AI group, as compared with 11.1 months (95% CI 9.9-13.6) in the first-line AI group with stratified log-rank test p-value of 0.0001 (Figure 3A). Similarly, median PFS in the second-line palbociclib + fulvestrant group was 10 months (95% CI 8.4-11.8), as compared with 5 months (95% CI 4.4-5.9) in the second-line fulvestrant group, with p-value < 0.0001 (Figure 3B). We further investigated and compared the dynamics of hazard rate of progression in each treatment group using estimated hazard functions. The difference in hazard rate curves between palbociclib and the control groups indicates that palbociclib treatment was associated with less early progression in both first- and second-line settings as compared to ET alone during the first year after treatment (Figure 3 C, D).

[Figure 3].

PFS curves and the hazard estimates over time according to treatments. The survival curves are shown up to maximum follow-up time of the CDK4/6 inhibitor treatment groups. (A) PFS1 survival curves and the shaded 95% CIs for the first-line setting. (B) PFS2 survival curves and the shaded 95% CIs for the second-line setting. (C) PFS1 hazard ratio curves are presented with shaded 95% CIs for first-line treatment cohort. (D) PFS2 hazard ratio curves are presented with shaded 95% CIs for the second-line treatment cohorts.

Median OS in the palbociclib + AI group was 44.3 months (95% CI 41.5-NA), as compared with 40.2 months (95% CI 37.9-44.2) in the control group and did not reach statistically significant p values in the first line setting (Figure 4A). In contrast, the addition of palbociclib to ET was associated with better OS in the second-line setting (Figure 4B). Specifically, the median OS in the palbociclib + fulvestrant group was 32.3 months (95% CI 28.1-37.4), as compared to 24.6 months (95% CI 21.8-27.7) in the second-line control group with stratified log-rank p-value of 0.0022 (Figure 4B). The hazard curves (Figure 4C and 4D) revealed that while in the first-line setting palbociclib treatment was not associated with changes in OS (Figure 4C), that in the second-line setting, for all the time intervals examined, palbociclib treatment was associated with longer OS (Figure 4D).

[Figure 4].

Overall survival curves and hazard estimates over time according to treatments. The survival curves are shown up to maximum follow-up time of the palbociclib treatment groups. (A) Survival curves and the shaded 95% CIs for the first-line setting. (B) Survival curves and the shaded 95% CIs for the second-line setting. (C) Hazard curves with shaded 95% CIs for the first-line setting. (D) Hazard curves with shaded 95% CIs for the second-line setting.

We also performed the inverse probability of treatment weighting (IPTW) analysis compared with the propensity score matching (Table S3 and Supplemental methods) , while the HR were similar between the two methods, the estimates that are generated using propensity matching method are more statistically robust. In particular, propensity matching method is more high-powered and efficient than IPTW method when the sample size of the treatment cohort is much smaller than the control cohort (i.e. 3 fold difference in the first-line analysis). Therefore, we base our main conclusion on the estimates from the propensity matching method.

We next assessed if palbociclib has OS benefit in the “long-term responders” in the first- and second-line settings. We defined long-term responders as those patients whose PFS times were longer than the median (i.e., 17.4 months for the first line and 10 months for the second line patients). The 17.4 and 10 months fixed times were selected as a landmark ^53,54 to conduct OS comparison between the two treatment groups (ET and palbociclib + ET). In the long-term responder subgroups, the patients had to be progression-free at the landmark times to continue the treatment interventions being examined. The demographic and clinical characteristics were very similar between the ET alone and palbociclib + ET long-term responder cohorts in both the first- and second-line settings (Table S4). Based on the logrank test of landmark analysis, there was no residual OS (measured from the landmark time point to death or censoring) difference between the AI alone and palbociclib + AI groups (p-value=0.5007) among first line therapy patients who were progression-free at 17.4 months (Fig S3A). However, among the second line therapy patients who were progression-free at 10 months, patients who received palbociclib + fulvestrant showed a longer overall survival than those with fulvestrant alone with p-value of 0.006 (Fig S3B).

Subgroup analyses support the better outcomes associated with palbociclib treatment across all of the subgroups in both first- and second-line settings for PFS and in the second-line setting for OS. (Figures 5A and 5B). The subgroup analysis of visceral vs. non-visceral was also performed under the Cox-proportional hazards models for PFS and OS (Figure 5). Compared with the PALOMA-3, our second-line cohort included 168 and 160 visceral metastases from the 380 control and treatment matched pairs, respectively. The estimated HRs were 0.49 (0.33-0.74) and 0.53 (0.31-0.9) for non-visceral and visceral metastasis, which are consistent with those reported for PALOMA-3 ^{9,11, 25–27}.

[Figure 5].

A: Hazard ratios (HRs) with shaded 95% confidence intervals for PFS and OS according to baseline characteristics in the first-line treatment analysis. HRs for subgroups with small sample sizes are not reported. B: HRs with shaded 95% confidence intervals for PFS and OS according to baseline characteristics in the second-line analysis. HRs for subgroups with small sample sizes are not reported.

Of note, additional lines of therapy (post palbociclib) or dose of fulvestrant in the second-line therapy favored better palbociclib treatment outcome without any suggestion of differential benefit among these categories (Figure 5B). Collectively, these results underscore that the risk of disease progression was significantly lower in the palbociclib treated group compared to the control group with HR+ disease in both first- and second-line settings. Moreover, the risk of death was also significantly lower in the palbociclib treatment group compared with the control group in the second-line setting in patients.

DISCUSSION

In this large, single institution study of patients with HR-positive breast cancer, palbociclib treatment exhibited a longer PFS when compared to patients receiving ET alone in the first- and second-line settings. Though the addition of palbociclib did not result in a statistically significant OS benefit in the first-line setting, we did find it significantly improved OS in patients treated with palbociclib compared to those receiving ET alone in the second-line setting. In the first-line setting, we observed an absolute difference of 6.3 months in median PFS with palbociclib + ET. In the second-line setting, we found an absolute difference of 5.0 months in median PFS with palbociclib treatment along with a significant 7.7 month longer OS compared to patients who received second-line fulvestrant only therapy. These findings from a large cohort of real-world patients, all from a single institution, not only show that patients receiving the combination of palbociclib + ET in the first- and second-line settings experience significantly longer PFS as compared to ET alone, but that palbociclib treated patients have an extended OS in the second-line setting.

Real-world data augments understanding of approved therapies beyond that learned from controlled clinical trials alone. Since clinical trials have strict eligibility criteria, real-world data provides insight into treatment of patients who might not have been eligible for the clinical trial leading to drug approval. Also, the majority of patients in the United States are not treated in the context of clinical trial eligibility and protocol requirements, so real-world data may be more representative of the effectiveness and tolerability of treatment in the general population. Therefore, our study extends our understanding of outcomes associated with palbociclib combined with ET.

Endocrine therapies are effective in both early-stage and metastatic HR+ breast cancer, providing a significant reduction in the risk of recurrence and death due to breast cancer. However the success of this treatment is limited by either intrinsic or acquired resistance ⁵³. The understanding of endocrine resistance mechanisms has fueled the development of strategies to prevent and/or overcome resistance by combining endocrine agents with drugs targeting several escape pathways, such as the CDK4/6 pathway ^{53, 54}.

In the current study, the median PFS of 17.4 months in the palbociclib + AI group, as compared with 11.1 months in the first-line AI group was shorter than those published in randomized controlled trials ^{10, 13, 17, 18} but similar to a real-world, retrospective analysis of electronic health records of the US healthcare system ³³. The first line PFS is shorter in the treatment cohort compared to PALOMA 2 as prognostic factors used in the analysis were different, and the number patients who presented with de-novo metastatic disease were higher in the PALOMA 2 trial, than in our cohort. Furthermore, the patients who we included in our cohort also included those referred to MD Anderson who had complete treatment information. However, these patients may have referral bias as the outcomes are population based and not representative of the institution with different demographic and clinical characteristics. However, the median PFS for the palbociclib + fulvestrant treatment group in our study of 10 months as compared to 5 months in the fulvestrant control group, is similar to the results from the PALOMA-3 trial ²⁴. Most of the patients enrolled on PALOMA-3 received 0 – 2 prior lines of systemic therapy for metastatic disease and approximately one-third received prior chemotherapy in the metastatic setting. It is important to note that our study included an ET treated group similar to the subgroup analysis in PALOMA-3 who had previously progressed after first-line endocrine therapy.

Our study did not show an OS benefit to palbociclib in the first line setting, in combination with AI, which is discrepant from other real-world analysis ³². The main different between our two studies that is likely to have impacted the results, is the sample size used. Our study started with a much larger sample size (Figure 1, N= 5402) than that of De Michele, et al ³² (n=1430), and as a result more patients were included in the propensity score matched cohorts in our study, which leads to more robust findings. Additionally, what sets our results apart from the PALOMA-3 trial ²⁴ is the significantly improved OS noted in the palbociclib treated patients, as compared to the matched historical controls. In our study the median OS of the palbociclib + fulvestrant group was 32.3 months as compared to 24.6 months in the second-line fulvestrant group. One reason for such significant differences in OS in our study cohort may be biological variance experienced due to previous and/or subsequent therapies. For example, in our cohort, the median age was 49 years, reflecting a younger population than in the PALOMA-3 (median age 57 years) ¹¹ trial. By assessing a large dataset of patients treated at a comprehensive cancer center, we found that there was no discrete timepoint among progressors that demonstrated benefit from palbociclib in the second-line setting, but there was a definite benefit in continuum throughout the duration at all timepoints for patients who received CDK4/6 inhibitors with fulvestrant compared to fulvestrant alone that complement the findings of the randomized clinical trials. Specifically, as shown in Figure 4D, the hazard rate associated with fulvestrant + palbociclib was significantly lower at all time intervals examined. The landmark OS analysis of the long-term responder cohorts further underscores that in the second line setting, patients who are treated with palbociclib have a significantly longer OS as compared to the fulvestrant alone arm. For these analyses, we used the median PFS of the propensity matched cohorts (Figure 3) as the landmark time ^{52, 55} and analyzed the subgroup of patients whose PFS time was longer than the calculated median. In other words, any subjects who died, progressed or was lost to follow-up prior to this designated time were excluded from the landmark analysis. The results of both the propensity matched patients (Figure 3) as well as the landmark analysis of the long-term responders (Figure S3) suggest that patients are more responsive to palbociclib + fulvestrant in the second line setting.

The addition of CDK 4/6 inhibitors has been promising in HR+/HER2− MBC. However, optimize use of CDK4/6 inhibitors are influenced by the differences among ribociclib, palbociclib and abemaciclib like pharmacokinetic factors, target selectivity, and toxicities ⁵⁶. Additional factors to consider when combining CDK4/6 inhibitors with AIs are the molecular mechanisms as to why treatment of patients with AI changes the biology of their tumors to one that can have a durable response to palbociclib-these mechanisms are currently not well understood. One possibility may be that AI itself could select for subpopulations, such those bearing ESR1 ⁵⁷ gene mutations that would render the patients less responsive to combination of AI + palbociclib and well-suited for palbociclib in combination with fulvestrant, an ET that can degrade ER and has in some, but not all studies shown evidence of better activity than AI in the setting of activating ESR1 mutations ⁵⁸

Some of the limitations of our study include it being a retrospective study and exclusion of heavily pretreated patients who received chemotherapy which does reflect the current real-world practice or multiple lines of endocrine therapy. Also, a majority of patients in the fulvestrant alone arm (61%) received a lower dose of fulvestrant (250 mg), which was the standard dose prior to the FDA approval of the 500 mg dose in 2010 on the basis of the CONFIRM Trial ⁵⁹ showing a benefit of 500 over 250 mg (OS HR 0.81); however, as shown on Fig 5B, the 500 mg subset showed significantly longer PFS and OS with the addition of palbociclib, almost identical to the entire “control” set with PFS HR for 500mg 0.56(0.39-0.81), all patients 0.51(041-0.64) and OS HR for 500 mg 0.62(0.41-0.94),all patients 0.67(0.52-0.87)”. The strengths of our study include the large sample size, all from one institution and long follow-up duration. This study is non contemporaneous with historical controls prior to CDK4/6 inhibitor approval, as number of patients not getting CDK4/6 inhibitors during the same period would be small and provide a bias. Another strength is the use of rigorous propensity score matching methods made possible by using a very large patient population to adjust for treatment selection biases, so that an “artificial” control group was created to approximate a quasi-randomized trial in reporting the real-world clinical outcomes. Our findings demonstrate that inclusion of palbociclib can lead to better disease control in both PFS and OS with AIs and fulvestrant.

Novelty and Impact:

We present real-world outcomes of a highly curated database accessing the largest population (n=5,402) yet reported of metastatic hormone receptor-positive and HER2-negative patients treated in the first- and second-line settings with palbociclib and endocrine therapy (ET), without confounding impacts of prior chemotherapy. Rigorous propensity score matching reveals palbociclib in combination with ET was associated with improved progression free survival in both first- and second-line settings and overall survival in the second-line setting compared with ET alone.

Abbreviations:

AI

Aromatase Inhibitor

AJCC

American Joint Committee on Cancer

ASCO/CAP

American Society of Clinical Oncology / College of American Pathologists

ASD

Absolute Standardized Difference

BC

Breast Cancer

CDKi

Cyclin-Dependent-Kinase-4/6 Inhibitor

Cis

Confidence Intervals

ER

Estrogen Receptor

ET

Endocrine Therapy

FISH

Fluorescence In Situ Hybridization

HR

Hormone Receptor

IHC

Immunohistochemistry

IRB

Institutional Review Board

MBC

Metastatic Breast Cancer

MDACC

MD Anderson Cancer Center

OS

Overall survival

PFS

Progression Free Survival

PgR

Progesterone Receptor

PH

Proportional Hazards

STROBE

Strengthening The Reporting of Observational studies in Epidemiology

Data Availability Statement:

The data that support the findings of this study are available on request from the corresponding author.