Real-world second- and third-line progression-free survival after progression on first-line CDK4/6 inhibitors in HR+/HER2- metastatic breast cancer by PAM50 intrinsic subtype: the SOLTI-1801 CDK-PREDICT study.

Introduction
Estrogen receptor-positive (ER+), HER2-negative (HER2-) metastatic breast cancer (MBC) constitutes the most common type of breast tumors and displays a heterogeneous clinical course driven by tumor hormone dependence [1]. First-line treatment with a cyclin-dependent kinase 4 and 6 inhibitor (CDK4/6i) in combination with endocrine therapy (ET) is the current standard of care, improving both progression-free survival (PFS) and overall survival (OS) [2–5]. However, resistance eventually develops, and the optimal therapy beyond first-line CDK4/6i remains a key clinical challenge.
In the second-line setting, treatment selection is increasingly guided by biological features and known resistance pathways. Available options include targeted agents like PI3K inhibitors (e.g., alpelisib for PIK3CA-mutant tumors), AKT inhibitors (e.g., capivasertib for tumors with PI3K/AKT pathway alterations), mTOR inhibitors (e.g., everolimus), or novel oral selective estrogen receptor degraders (SERDs) such as elacestrant for ESR1-mutated tumors. CDK4/6i continuation and combination therapies, such as imlunestrant plus abemaciclib [6–13], are valid options regardless of mutational status. Despite these advances, clinical outcomes remain variable, highlighting the need for biomarkers to guide post-CDK4/6i treatment decisions.
The PAM50 intrinsic subtypes (IS) (Luminal A (LumA), Luminal B (LumB), HER2-enriched (HER2E), Basal-like (BL), and Normal-like (NL)) have demonstrated clear prognostic value in HR+/HER2- MBC treated with ET plus CDK4/6i in the first-line setting, both in the MONALEESA trials and in the previous analysis of the CDK-PREDICT study [14–16]. While the BOLERO-2 trial also demonstrated the prognostic relevance of PAM50 IS in patients progressing on aromatase inhibitors, this biomarker remains largely unexplored in the current post-CDK4/6i setting [17]. Additionally, PDCD1 (the gene encoding PD-1) and CCNE1 have shown prognostic value in both the CDK-PREDICT study and the PALOMA trials [18, 19] and tumor-infiltrating lymphocytes (TILs ≥ 10%) were prognostic in the CDK-PREDICT study [16]. In particular, evidence remains limited on the prognostic impact of IS and other biomarkers derived from metastatic tissue in real-world second- and third-line settings, which now include a wide range of targeted agents.
To address this gap, we conducted a post hoc secondary analysis within the SOLTI-1801 CDK-PREDICT study to assess the prognostic impact of IS and additional transcriptomic biomarkers on real-world second-line progression-free survival (rwPFS-2L) and real-world third-line progression-free survival (rwPFS-3L) in patients with HR+/HER2- MBC previously treated with CDK4/6i. This analysis integrates real-world treatment data across both settings, offering insight into clinical and biological patterns after CDK4/6i treatment.

Methods

Study design
The SOLTI-1801 CDK-PREDICT study was a prospective, multicenter, observational cohort that enrolled patients with HR+/HER2- metastatic breast cancer treated with first-line CDK4/6i plus ET, with mandatory baseline metastatic biopsies for centralized PAM50 IS profiling. The original study demonstrated the prognostic value of PAM50 IS in the first-line setting and provides the biological and clinical framework for the present work.
The current study is a post hoc secondary analysis conducted within the previously published SOLTI-1801 CDK-PREDICT cohort. In this analysis, patients were longitudinally followed to characterize real-world treatment patterns and PFS outcomes in the second- and third-line settings and to explore associations with PAM50 IS and selected biomarkers.

Study population
Patients from the CDK-PREDICT study were eligible for this analysis. Key inclusion criteria included a biopsy-confirmed diagnosis of H+/HER2-MBC per ASCO/CAP guidelines, availability of a formalin-fixed paraffin-embedded (FFPE) metastatic biopsy suitable for gene expression profiling within 90 days prior to first-line CDK4/6i, ECOG performance status 0–2, and adequate organ function. Patients who discontinued CDK4/6i within the first 2 months for reasons other than progression were excluded. Clinical data were retrospectively collected and prospectively updated, including first-line PFS, OS, and treatment sequences after progression on CDK4/6i plus ET, through second- and third-line systemic therapy [16].

Study objectives
The primary objective was to evaluate the prognostic impact of baseline PAM50 IS (luminal vs non-luminal and by specific subtypes), assessed in metastatic tissue prior to CDK4/6i initiation, on rwPFS-2L and rwPFS-3L in this homogeneous cohort. Secondary objectives were (1) to describe real-world systemic treatments administered in the second and third lines within the CDK-PREDICT cohort, including mutational status when available; and (2) to explore rwPFS-2L associations with additional biomarkers—such as CCNE1 and PDCD1 expression and TILs—analyzed as continuous and categorical variables using predefined, previously published cutoffs in the CDK-PREDICT study [16]. An exploratory analysis also evaluated rwPFS-2L and rwPFS-3L by type of therapy received, given the heterogeneity and small sample sizes of treatment subgroups; results are reported in the Supplementary Appendix.

Outcomes
The primary outcomes of this analysis were rwPFS-2L and rwPFS-3L, defined as the time from initiation of the respective line of systemic therapy to investigator-assessed disease progression or death from any cause, whichever occurred first. Patients without a documented event were censored at the date of the last clinical assessment.

Variables and molecular analyses
Therapeutic groups in the second and third lines were categorized as follows: (1) ET monotherapy (tamoxifen, aromatase inhibitors, fulvestrant); (2) targeted therapies, alone or in combination (oral SERDs, PI3K or PARP inhibitors, among others); (3) capecitabine; (4) other chemotherapies; and (5) antibody–drug conjugates (ADCs). Data collected included treatment type, start date, and progression date. Molecular data were derived from baseline metastatic FFPE biopsies collected in the CDK-PREDICT study [16], centrally reviewed to ensure ≥ 10% tumor content. Histopathological markers (Ki-67, PR, TILs) and gene expression profiling (nCounter, NanoString Technologies) included PAM50 classification and expression of cell cycle and immune-related genes. Gene expression data were normalized, log2-transformed, and PAM50 IS assigned using the published algorithm. Biomarkers were analyzed as continuous or categorical variables with predefined cutoffs [16]. Somatic mutation data from routine testing (multigene panels, single-gene assays, or liquid biopsy) were collected when available. No additional molecular testing for PIK3CA or ESR1 mutations was performed as part of the study protocol. Patients with known germline BRCA1/2 mutations identified through routine clinical testing were included, and PARP inhibitors were administered as part of standard-of-care treatment when clinically indicated.

Statistical analysis
Descriptive statistics summarized baseline characteristics by line of therapy and treatment regimen in the second- and third-line settings. Kaplan–Meier methods estimated rwPFS-2L and rwPFS-3L across subgroups. The prognostic value of PAM50 IS was assessed using univariable and multivariable Cox proportional hazards models adjusted for clinical covariates (de novo metastatic disease, endocrine sensitivity (as relapse ≥ 12 months after completion of adjuvant ET) in accordance with ESMO ABC5 [20], visceral involvement, prior CDK4/6i duration < 12 vs ≥ 12 months) and treatment type in the corresponding line. Unless otherwise specified, hazard ratios (HR) reported in multivariable analyses correspond to Cox models adjusted for the covariates described. Clinical data were reviewed for completeness and consistency prior to analysis. Lines of systemic therapy were defined sequentially based on documented disease progression and treatment changes in routine clinical practice. All analyses and plots were performed using STATA version 15.1 and R software.

Ethics
The study complied with Good Clinical Practice and the Declaration of Helsinki; all patients gave written informed consent, and institutional review boards approved the protocol (Ref: 07/152297.9/21).

Results
From May 2020 to May 2022, 185 patients with available clinical and genomic data (PAM50) from six Spanish hospitals were included in the CDK-PREDICT study. Follow-up was subsequently extended until February 2025. The updated median follow-up was 59.3 months (95% CI 52.6–65.9). In the updated first-line analysis, median PFS was 32.7 months in luminal subtypes and 10.0 months in non-luminal subtypes (HR = 2.71; 95% CI 1.78–4.14; p < 0.05; reference: luminal). Likewise, updated median OS was 74.5 vs 36.2 months (HR = 2.43; 95% CI 1.48–3.98; p < 0.05; reference: luminal).
Of the 185 patients included in the CDK-PREDICT study, 48 (26%) were still on first-line therapy at data cutoff, 11 (6%) died before starting second-line treatment, and 1 (< 1%) did not receive further therapy for other reasons, resulting in 125 evaluable patients for rwPFS-2L analysis. Among these, 16 (13%) were still on second-line therapy and 14 (11%) died before initiating third-line treatment, resulting in 95 patients evaluable for rwPFS-3L analysis (Fig. 1 and Supplementary Fig. 1).
In the overall cohort (n = 185), LumA and LumB subtypes predominated (39% and 44%, respectively), while HER2E, BL, and NL were less frequent. Across subsequent therapy lines, a progressive decrease in LumA and a relative increase in LumB and HER2E were observed (Fig. 1). This decline in LumA across treatment lines was statistically significant (Cochran–Armitage trend test, p < 0.01). Importantly, these shifts reflect population-level selection rather than true biological subtype switching, as PAM50 profiling was performed only at baseline and no systematic rebiopsy was conducted at progression. Median age was similar across cohorts—61 years in the overall and rwPFS-2L population, and 59 years in rwPFS-3L. The most frequently used first-line CDK4/6i were palbociclib (46.5%) and ribociclib (44.4%), followed by abemaciclib (9.1%). Most patients were postmenopausal at diagnosis (76.2%) and had endocrine-sensitive disease (77.8%). Table 1 summarizes baseline characteristics for the overall population and for patients receiving second- and third-line therapy.

Real-world PFS-2L and rwPFS-3L by PAM50 Intrinsic Subtypes
The median rwPFS-2L was 7.2 months for luminal subtypes versus 6.1 months for non-luminal subtypes (HR 1.40; 95% CI, 0.86–2.30; p = 0.180). By individual subtype, median rwPFS-2L was 6.4 months for LumA (reference), 7.6 for LumB, 6.2 for HER2E, 3.0 for BL, and 9.0 for NL. Only the BL subtype showed a significantly shorter median rwPFS-2L (HR 3.82; 95% CI, 1.07—13.63; p = 0.020). Table 2 displays the complete univariable and multivariable analyses of rwPFS-2L.

In the third-line setting, median rwPFS-3L was 6.4 months for luminal versus 3.3 months for non-luminal subtypes (HR 1.74; 95% CI 0.98–3.08; p = 0.058; reference: luminal). Among subtypes, only BL was significantly associated with shorter median rwPFS-3L (HR 5.63; 95% CI 1.17–27.02; p = 0.031) (Fig. 2). rwPFS-2L and rwPFS-3L by all PAM50 IS, including HR with 95% CI and p-values, are provided in Supplementary Table 1A–B.

Real-World 2nd- and 3rd-Line Treatments Used in the CDK-PREDICT Study and Mutational Status
In the second-line setting (n = 125), real-world therapeutic strategies included endocrine monotherapy (22%; n = 27), targeted therapies (51%; n = 65), capecitabine (10%; n = 12), and other chemotherapy agents (17%; n = 21). The most frequently detected somatic mutation (per local standard of care) was PIK3CA, identified in 34% of cases (n = 42), followed by ESR1 mutations in 14% (n = 18). Missing data for mutational status were substantial—27% for PIK3CA (n = 34) and 42% for ESR1 (n = 53). Complete mutational data for all analyzed genes are provided in Supplementary Table 4. The timing of mutational testing (baseline vs post-progression) was heterogeneous and not systematically recorded. In the third-line setting (n = 95), treatment options were more heterogeneous, including endocrine monotherapy (2%; n = 2), targeted therapies (29%; n = 27), capecitabine (36%; n = 34), other chemotherapy (26%; n = 25), and ADCs (7%; n = 7). The specific distribution by treatment subgroups, including PI3K inhibitors, mTOR inhibitors, and oral SERDs, is shown in Fig. 3.

Additional analyses of biomarkers in rwPFS-2L
We explored the association of additional biomarkers with rwPFS-2L, including TILs, CCNE1, and PDCD1 expression (Fig. 4). TILs analyzed as a continuous variable were not significantly associated with rwPFS-2L (HR 1.01; 95% CI 0.98–1.04; p = 0.437). Patients with TILs ≥ 10% had a numerically shorter median rwPFS-2L than those with TILs < 10% (3.5 vs. 7.1 months), and multivariable analysis showed a non-significant trend toward shorter outcomes (HR 1.12; 95% CI 0.56–2.23; p = 0.750). CCNE1 expression was significantly associated with worse rwPFS-2L as a continuous variable (HR 1.22; 95% CI 1.02–1.47; p = 0.033), and patients with high expression (≥ 25th percentile) had shorter median rwPFS-2L than those with low expression (6.1 vs. 9.0 months). In multivariable analysis, this association was attenuated (HR 1.39; 95% CI 0.83–2.33; p = 0.209). PDCD1 expression was not significantly associated with rwPFS-2L (HR 1.01; 95% CI 0.94–1.29; p = 0.246); patients with low expression (< 75th percentile) had a numerically longer median rwPFS-2L than those with high expression (7.2 vs. 6.4 months; adjusted HR 1.10; 95% CI 0.94–1.29; p = 0.246).

Real-world PFS-2L and PFS 3L according to therapeutic subgroups
In the exploratory analysis of rwPFS-2L by therapeutic group, no statistically significant differences were observed. Median rwPFS-2L was 5.0 months for endocrine monotherapy (n = 27), 7.0 months for targeted therapies (n = 65), 7.4 months for capecitabine (n = 12), and 8.7 months for other chemotherapy agents (n = 21). Although not statistically significant, endocrine monotherapy showed the shortest median rwPFS-2L. Among specific targeted agents, PI3Ki (n = 18) achieved the longest median rwPFS-2L at 12.0 months, followed by CDKi (n = 8; 10.9 months) and other single-agent chemotherapies (n = 21; 8.0 months). In the third-line setting (n = 95), a similar trend was observed: endocrine monotherapy (n = 2) had the shortest median rwPFS-3L (3.3 months), while targeted therapies (n = 27), capecitabine (n = 34), and other chemotherapies (n = 25) achieved 6.0, 6.9, and 5.4 months, respectively. ADCs (n = 7) achieved the longest median rwPFS-3L at 9.9 months, showing a non-significant trend toward improved outcomes versus endocrine monotherapy (HR 0.19; 95% CI 0.03–1.09; p = 0.063). Supplementary Fig. 2 shows Kaplan–Meier curves for rwPFS-2L and rwPFS-3L, and Supplementary Tables 2A–B and 3A–B detail aggregated and subgroup data.

Discussion
Our study provides one of the most comprehensive real-world assessments to date of clinical outcomes and transcriptomic biomarkers in patients with HR+/HER2- MBC treated beyond first-line CDK4/6 inhibition. Several relevant findings emerge from this secondary analysis of the CDK-PREDICT cohort, which help to better characterize prognostic factors and therapeutic dynamics in the post-CDK4/6i setting.
We demonstrated the prognostic relevance of PAM50 IS beyond first-line CDK4/6i. The clinical and prognostic utility of these subtypes had been previously shown in randomized trials such as MONALEESA, in the initial analysis of the CDK-PREDICT study, and in the pre-CDK4/6i era BOLERO-2 trial [14, 16, 17]. Our findings further extend this evidence to later treatment lines. Across both second- and third-line settings, BL tumors were associated with significantly shorter rwPFS, underscoring their aggressive phenotype despite hormone receptor positivity. Interestingly, the proportion of LumA tumors declined in second and third lines, while more proliferative subtypes such as LumB and HER2E became more frequent. These percentage changes are explained by the predominance of LumA tumors among patients who had not progressed on first-line therapy at data cutoff, reflecting greater hormone sensitivity and a more indolent disease course. By contrast, a single study analyzing 39 paired biopsies reported true IS switching in 61.9% of cases, with HER2E tumors increasing from 36 to 51% [21]. In that cohort, the prognostic impact of the BL subtype was particularly marked, although the small sample size limits the strength of the conclusions. However, unlike that study, our analysis reflects not biological shifts but the selective retention of LumA tumors in earlier lines. Consequently, the PFS-2L and PFS-3L populations were proportionally enriched with biologically more aggressive subtypes. Although differences in rwPFS-2L and rwPFS-3L between luminal and non-luminal subtypes did not reach statistical significance, the effect sizes observed—particularly in third line—suggest meaningful clinical divergence and warrant further investigation.
Therapeutic strategies after CDK4/6i progression were heterogeneous, mirroring real-world clinical practice where treatment decisions are influenced by multiple factors, including biological markers, prior response, tolerability, and drug availability. In a large United States electronic health record–based cohort (n = 1210), Martin et al. reported that after CDK4/6i progression, 36% of patients received another CDK4/6i, 29.7% switched to chemotherapy and 11.7% received everolimus combinations, with continuation of CDK4/6i being associated with superior rwPFS (HR 0.48) and OS (HR 0.30) compared with chemotherapy [22]. A systematic literature review of 18 real-world studies highlighted the lack of an established standard after CDK4/6i progression, showing that the effectiveness of available second-line regimens remains limited. Weighted median rwPFS was only 3.9 months with single-agent ET, 3.6 months with mTOR inhibitors ± ET, and 6.1 months with single-agent chemotherapy, reflecting the modest benefit of current strategies [23]. In line with these findings, a single-institution retrospective cohort of 140 patients with HR+/HER2-MBC reported a median PFS of just 6.0 months following progression on CDK4/6i, underscoring the poor prognosis beyond first-line therapy [24].
Recent randomized trials have further informed treatment strategies beyond progression. In the phase II MAINTAIN trial, switching ET while continuing CDK4/6 blockade with ribociclib improved PFS to 5.3 months versus 2.8 months with ET alone (HR 0.57; 95% CI, 0.39 to 0.85) [11]. Similarly, the phase III postMONARCH trial showed a modest but significant benefit with abemaciclib plus fulvestrant (median PFS 6.0 vs 5.3 months; HR 0.73; 95% CI 0.57 to 0.95) [12]. In EMBER-3, the oral SERD imlunestrant improved PFS over standard ET in both ESR1-mutant and wild-type tumors, with the greatest benefit observed in ESR1-mutant disease, and the combination of imlunestrant plus abemaciclib further prolonged PFS (9.4 vs 5.5 months; HR 0.57; 95% CI 0.44 to 0.73) [13] regardless of ESR1 mutation status. Moreover, the recently published REIGNITE meta-analysis, pooling five randomized trials (n = 1184), demonstrated that continuing or switching CDK4/6 inhibition after first-line progression significantly improved PFS compared with ET alone (HR 0.73), with consistent benefit observed in patients harboring PIK3CA or ESR1 mutations [25]. Real-world data also support the feasibility of everolimus plus ET after CDK4/6i progression [9].
Taken together, these data illustrate how post-CDK4/6i treatment decisions are increasingly shaped by both biomarker-guided strategies (e.g., ESR1 and PI3K pathway alterations) and biomarker-agnostic approaches. The CDK-PREDICT cohort reflects a contemporary real-world population treated during a period of rapidly evolving therapeutic options, in which some newer agents were emerging but not yet uniformly adopted. At the time of enrollment, adjuvant CDK4/6i were not part of standard clinical practice; therefore, our findings primarily reflect outcomes in patients treated with CDK4/6i exclusively in the metastatic setting. Despite these evolving treatment paradigms, the observed patterns remain informative for understanding outcomes and biomarker associations beyond first-line CDK4/6 inhibition, while future studies will be needed to assess the impact of earlier CDK4/6i exposure.
Importantly, real-world PFS estimates should not be interpreted as directly comparable to PFS reported in randomized clinical trials, given differences in patient selection, assessment schedules, censoring rules, and treatment discontinuation criteria. Therefore, trial data are discussed here for contextualization rather than quantitative comparison. Endocrine monotherapy was consistently associated with the shortest rwPFS across lines, reinforcing its limited role in this setting. This pattern is also evident in recent second-line clinical trials—such as EMERALD, CAPItello-291, and VERITAC-2—where fulvestrant or aromatase inhibitors achieved only 1.8–3.7 months of PFS, compared with novel agents like elacestrant, capivasertib, or vepdegestrant [10, 13, 26]. In this context, PI3K pathway inhibition has shown clear clinical benefit. In the SOLAR-1 trial alpelisib plus fulvestrant achieved a median PFS of 10.9 months versus 3.7 months with fulvestrant alone (HR 0.61; 95% CI, 0.50 to 0.85) [6] in a population mostly CDK4/6i-naïve, whereas in the BYLieve study, conducted in patients previously treated with CDK4/6i, a median PFS of 7.3 months was reported [27]. Similarly, in CAPItello-291, capivasertib plus fulvestrant improved median PFS to 7.2 months versus 3.6 months in the overall population (HR 0.60; 95% CI 0.51 to 0.71), and to 7.3 versus 3.1 months (HR 0.50; 95% CI, 0.38 to 0.65) in patients with PIK3CA/AKT1/PTEN alterations [7]. Recent data from the EVERA and VIKTORIA trials presented at ESMO 2025 further support this approach, showing encouraging PFS benefits with new PI3K/AKT inhibitors after CDK4/6i [28, 29]. These findings are directionally consistent with real-world observations, where patients receiving PI3K pathway inhibitors in later lines derive clinically meaningful benefit, although rwPFS estimates cannot be directly compared with trial-based PFS outcomes. Promising early results are also emerging with next-generation agents, such as RLY-2608, further supporting the therapeutic potential of targeting the PI3K/AKT pathway beyond endocrine monotherapy in this setting.
In the third line, ADCs demonstrated the most favorable outcomes in our cohort. Although based on a limited number of patients, this observation aligns with emerging evidence supporting the role of ADCs beyond conventional chemotherapy [31–33]. Importantly, non-luminal subtypes—present in around 25% of patients—may help identify those who could benefit more from early use of ADCs. Patients with early progression during the first 6–12 months of CDK4/6i therapy could be prioritized for this strategy, highlighting the need for confirmation in larger prospective studies.
With respect to somatic DNA alterations, it should be noted that PIK3CA and ESR1, among other genes, were not pre-specified study objectives and no protocol-mandated genomic testing was performed; therefore, mutational data were collected only when available from routine clinical practice. In this context, PIK3CA mutations were detected in 34% of evaluable patients, consistent with the reported prevalence of ~ 35–45% in HR+/HER2- breast cancer [6, 34, 35]. ESR1 mutations were detected in 14% of cases, lower than the ~ 20–40% typically seen after aromatase inhibitors and the ~ 30–50% reported post-CDK4/6i exposure [10, 36, 37]. This likely reflects real-world limitations: missing mutation data in a relevant proportion of patients (27% for PIK3CA and 42% for ESR1), non-standardized timing of ESR1 testing—frequently performed before progression on aromatase inhibitors—and heterogeneous testing methodologies across centers. Together, these factors may have contributed to under-detection of acquired ESR1 mutations. Notably, large genomic studies consistently identify PIK3CA and ESR1 as the most frequent somatic alterations following CDK4/6i plus ET [37], underscoring the importance of systematic and timely genomic profiling to inform post-CDK4/6i treatment decisions.
Exploratory analyses of CCNE1, PDCD1, and TILs revealed biomarker patterns in line with prior evidence. In this exploratory context, high CCNE1 expression was associated with numerically shorter rwPFS, consistent with previous reports linking CCNE1 overexpression to endocrine resistance, reduced sensitivity to CDK4/6 inhibition, and poorer outcomes in HR+/HER2-disease [18, 38]. Elevated PDCD1 expression was likewise associated with poorer outcomes in our cohort. In line with recent findings in luminal ER+/HER2-breast cancer, high TIL density appears to reflect a more proliferative and aggressive tumor biology, correlating with adverse clinicopathologic features and worse breast cancer-specific survival rather than effective immune control [39, 40]. In the previously published CDK-PREDICT cohort, high CCNE1, PDCD1, and high TIL density were prognostic even in the first-line CDK4/6i therapy. These findings are supported by translational analyses from PALOMA-3, where high CCNE1 predicted lack of benefit from palbociclib [38] and by the PEARL study, which showed that non-luminal tumors or those with CCNE1 overexpression derived greater benefit from chemotherapy than from CDK4/6i plus ET [18]. Altogether, these results should be interpreted as hypothesis generating, suggesting that transcriptomic and immune biomarkers may help guide earlier use of chemotherapy or novel combinations in patients less likely to benefit from ET.
Overall, our results highlight the value of integrating molecular subtyping and biomarker profiling after CDK4/6i therapy. The expanding availability of novel ER-targeted and PI3K/AKT inhibitors offers new endocrine-based options, while ADCs are emerging as effective alternatives to chemotherapy. However, the optimal treatment sequence remains unclear. PAM50 IS and selected biomarkers may help guide therapy, particularly in non-luminal or endocrine-resistant disease. Future studies should incorporate PAM50 and genomic markers such as CCNE1 and TILs to refine sequencing strategies and improve outcomes in the post-CDK4/6i setting.
This study has several limitations. First, its observational and real-world nature precludes causal inference and introduces potential selection bias, while treatment decisions were not randomized and may have been influenced by unmeasured factors. Additionally, the requirement for a baseline metastatic biopsy suitable for transcriptomic profiling may have selected for patients with more accessible disease and better performance status, potentially limiting generalizability. Second, sample sizes were relatively modest—especially in the third-line group and within specific treatment subgroups—limiting statistical power. Third, molecular analyses relied on baseline metastatic biopsies without serial sampling, preventing assessment of temporal tumor evolution and potential PAM50 switching. In addition, mutation testing was not uniformly performed; timing and methodology varied across centers, reflecting routine clinical practice rather than protocol-mandated testing, which likely led to underestimation of ESR1 prevalence and variability in PIK3CA reporting. Moreover, the post hoc nature of this secondary analysis introduces potential residual confounding despite multivariable adjustment. Finally, toxicity and quality-of-life outcomes were not captured. Despite these caveats, our study provides valuable insights into prognostic stratification and treatment patterns in a modern real-world cohort of HR+/HER2-MBC beyond CDK4/6i, largely treated in academic centers, and supports further investigation of transcriptomic and molecular biomarkers to guide individualized care in later treatment lines.

Conclusion
Treatment outcomes after CDK4/6i progression differ across PAM50 IS. Integrating molecular subtyping and biomarker profiling into clinical practice may help guide therapy selection and optimize management beyond the first-line treatment.

Supplementary Information
Below is the link to the electronic supplementary material.