Article selection
A total of 8,963 records were identified using different databases. 2,284 records were excluded because of duplication, and 6,564 records were excluded after the initial title and abstract screening due to unmet inclusion criteria. Of the remaining 115 records, the full-text articles were carefully read, and 14 records were excluded due to insufficient pCR data and inaccessibility. Finally, 101 studies fulfilled the eligibility criteria and were included in the systematic review and meta-analysis (Fig. 1).
Overview of included studies
Altogether, 19,708 Asian breast cancer patients were gathered from the 101 studies, with an average of 195 patients per study (see Supplementary Table 1, Additional File 4). The study population comprised 91 studies from Eastern Asia (China [12, 34–116], Hong Kong [117, 118], Korea [119], and Japan [120–123]), 7 studies from Western Asia (Egypt [124], Iran [125], Turkey [126, 127], and Saudi Arabia [128–130]), 2 studies from Southeast Asia (Indonesia [131], Singapore and Malaysia [132]), and 1 study from South Asia (India [133]). The recruitment period for the patients enrolled in the studies ranged from 1991 to 2020. Most of the study cohorts followed hospital-based study design (76.2%) and clinical trials (14.9%), while 8.9% were not reported. 61.3% of the hospital-based study designs were conducted in unicenter, while 14.9% were in multicenter. Of the 101 studies, only 65 studies provided molecular subtypes data (n = 11361) - TNBC (29.2%), HER2E (20.7%), luminal B (13.2%), and luminal A (7.2%), while 1.6% were missing. Several studies did not categorise their luminal subtype into luminal A or B. Hence, they were reported as luminal-like (24.9%) in this study. Only 69 studies provided data on biomarkers, comprising the routinely analysed biomarkers - ER (25.1%), PR (25.1%), HR (5.7%), HER2 (27.4%), and Ki-67 (15.5%) – and several non-conventional biomarkers investigated specifically for the study including EGFR (11.0%), CK5/6 (8.0%), Tau (11.7%), Androgen Receptor (AR) (8.6%), PDL1 (3.6%), P-glycoprotein (P-gp) (3.3%), DNA topoisomerase II-alpha (Topo IIa) (3.2%), p53 (9.8%), and others (0.8-4.0%) (see Supplementary Table 1, Additional File 3). Meanwhile, 7 studies provided information on genetic variations and differential expression, where the common genetic variations reported were from TP53 (15.7%), PIK3CA (24.6%), MYC (5.8%), ERRB2 (5.8%), CCDN1 (5.8%), BRCA1 (15.7%), BRCA2 (15.7%) and others (10.9%).
All the patients in the studies received NAC, and 25.0% of patients received follow-up adjuvant therapy. Collectively, NAC taxane-anthracycline (TA), taxane-platinum (TP), and taxane-anthracycline-platinum (TAP) combination were mentioned in 41.5%, 18.6%, and 1.5% of the studies, respectively. Meanwhile, 18.3% were treated with NAC anthracycline-based chemotherapy, 14.0% were treated with NAC taxane-based chemotherapy, and 8 studies did not provide specific treatment information. Some patients were treated concomitantly with targeted therapy and endotherapy (23.0%). The definition of pCR used in the included studies was mostly not reported according to any guideline (71.3%), with only 28.7% reported pCR following the Miller-Payne grading (17.8%), Kuerer et al. (1.9%), RECIST (2.9%) and other grading systems (6.1%) including the Ribero classification, Japanese Breast Cancer Society v2007, USFDA guideline, WHO criteria and pathological TNM system.
Furthermore, 33 of the included studies conducted the multivariate analysis. Most of the included variables used to adjust the multivariate analysis were the commonly reported biomarkers (ER, PR, HER2, and Ki-67), age, tumour size and grade, age at diagnosis, lymph node stage, histological grade, body mass index (BMI), chemotherapy regimens, chemotherapy cycles, and other biomarkers and genetic variations unique to the study.
Quality of the included studies
The quality assessment of the studies was presented in Supplementary Tables 1 and 2 (see Additional File 6). Fourteen (14) case-control studies and 87 cohort studies were included in our systematic review. NOS scores for the 87 cohort studies ranged from 5 to 9 stars, and NOS scores for the 14 case-control studies ranged from 4 to 9 stars. No study was excluded since all studies scored ≥ 4 stars.
A summary of the risk of bias assessed on each question using the NOS for cohort and case-control studies is shown in Supplementary Figs. 1 and 2(see Additional File 6). We considered both breast cancer treatment and characterisation as the most important factors for adjustment in the comparability domain because our study eligibility criteria required adjustment for the involvement of somatic genetic polymorphisms or biomarkers or molecular subtypes in breast cancer treatment response. Following this consideration, only 33% (n = 29/87) cohort studies earned a star for comparability regarding pCR and breast cancer treatment, while 94% (n = 82/87) cohort studies earned a star regarding pCR and breast cancer characterisation. As for case-control studies, 79% (n = 11/14) studies earned a star for comparability regarding pCR and breast cancer treatment, while 93% (n = 13/14) studies earned a star regarding pCR and breast cancer characterisation. Notably, when both adjustment factors were combined, only 31% (n = 27/87) cohort studies and 71% (n = 10/14) case-control studies earned both stars in the comparability domain. Among the cohort studies, evaluation of the selection of the non-exposed cohort was the question with the lowest count of stars, with only 32% (n = 28/87) of the studies having a low risk of bias. Meanwhile, among the case-control studies, apart from the first adjustment in the comparability domain, the lowest count of stars was for the question evaluating the selection of controls, with 79% (n = 11/14) of studies showing a low risk of bias.
Association of breast cancer characterisation and treatment response
The molecular subtypes classification of breast cancer, presence or absence of specific biomarkers, and genetic variations in the breast cancer diagnosis can be utilised to predict the pCR outcome in patients treated with specific chemotherapeutic agents. Five molecular subtypes, fourteen biomarkers, and eleven genetic variations were qualitatively evaluated for their predictive value in Asian breast cancer patients (see Supplementary Table 2.2, Additional File 4). Meanwhile, of the 101 studies, 60 studies provided data that could be used for meta-analysis (Figs. 2–7 and Supplementary Figs. 1–8, Additional File 7). All the qualitative and meta-analyses results are presented by the molecular classification, biomarkers, and genetic characterisation of Asian breast cancer patients. Additionally, meta-analysis results using the Mantel-Haenszel method listed under each breast cancer characteristic are grouped according to the chemotherapeutic agents. Meta-analysis results using the inverse-variance method are presented separately since it is pooling the reported association data, for which they are presented by the breast cancer characteristics as well.
Molecular subtypes classification
Qualitatively, most studies with molecular subtypes classification provide data on the pCR rates of patients treated with TA and TP regimens. When treated with TP, luminal A had the lowest pCR rate of 7.7% (n = 4), while the highest pCR rate was observed in HER2E at 52.4% (n = 7). In comparison, the pCR rates in the other subtypes were 32.5% (n = 3), 28.1% (n = 4) and 41.7% (n = 7) for luminal-like, luminal B, and TNBC, respectively. However, when treated with NAC TA, the highest pCR rate was observed in TNBC at 30.4% (n = 20), and similarly, the lowest pCR rate was observed in luminal A at 4.3% (n = 5). Comparatively, the pCR rates in luminal-like, luminal B, and HER2E were 9% (n = 6), 12% (n = 5), and 27.8% (n = 9), respectively. Our findings suggest that patients with TNBC and HER2E subtypes treated with NAC TP and TA were more likely to obtain pCR, while luminal A was less likely to obtain pCR with both regimens.
Under meta-analysis, the role of molecular subtypes was examined in NAC TA-treated (Fig. 2 and Supplementary Fig. 1, Additional File 7) and TP-treated (Figs. 3–4 and Supplementary Fig. 1, Additional File 7) breast cancer patients.
Taxane-anthracycline (TA) chemotherapy
Our study first compares the effect of HER2E and luminal subtypes on pCR outcomes in TA-treated patients (Fig. 2A-D). In the analysis of four pooled studies [39, 48, 51, 71] comparing HER2E and luminal-like subtypes, 494 were identified with HER2E subtype and 764 patients with luminal-like subtype. Despite the large number of patients with luminal-like subtypes, our findings significantly associate patients with HER2E subtypes with better pCR outcomes (OR: 4.08; 95% CI; 2.78-6.00; p < 0.0001; Fig. 2A). When HER2E patients were compared with patients with luminal A subtype, patients with HER2E subtypes were also found to be significantly more likely to achieve pCR (OR: 5.27; 95% CI; 1.16–23.86; p = 0.03; Fig. 2B). Similarly, when HER2E was analysed against luminal B, HER2E was significantly associated with pCR (OR: 2.78; 95% CI: 1.42–5.44; p = 0.003; Fig. 2C). To confirm and elucidate the effect of luminal subtypes on pCR outcome in NAC TA-treated breast cancer patients, we combined all luminal data (luminal-like, luminal A, and luminal B) as luminal, combined and compared it against HER2E patients. Nine studies [37, 39, 48, 51, 60, 61, 68, 71, 110] were pooled, yet our findings still showed that HER2E subtype was significantly associated with pCR outcome in NAC TA-treated patients (OR: 3.89; 95% CI: 2.69–5.64; p < 0.0001; Fig. 2D).
Similarly, analyses conducted comparing the effect of TNBC and luminal subtypes on pCR outcome in TA-treated patients (Fig. 2E-H) significantly associate TNBC with better pCR outcome compared to luminal-like (OR: 4.45; 95% CI: 2.79–7.11; p < 0.0001; Fig. 2E), luminal A (OR: 11.66; 95% CI: 3.64–37.38; p < 0.0001; Fig. 2F), luminal B (OR: 3.89; 95% CI: 2.20–6.87; p < 0.0001; Fig. 2G), and luminal, combined (OR: 4.59; 95% CI: 3.35–6.29; p < 0.0001; Fig. 2H).
To further explore the effect of molecular subtypes on the pCR outcome of NAC TA-treated breast cancer patients, the TNBC subtype was compared to the HER2E subtype revealing that neither was associated with pCR (OR: 1.17; 95% CI: 0.80–1.70; p = 0.43; Supplementary Fig. 1A, Additional File 7). Similarly, an analysis between luminal A and luminal B showed that neither was associated with better pCR outcome when treated with NAC TA (OR: 2.47; 95% CI: 0.79–7.73; p = 0.12; Supplementary Fig. 1B, Additional File 7).
Taxane-platinum (TP) chemotherapy
In the analysis pooling three studies [45, 64, 102] comparing HER2E and luminal-like subtypes, 321 patients were identified with HER2E subtype, and 382 patients were luminal-like. When treated with NAC TP, neither HER2E nor luminal-like patients were associated with pCR (OR: 1.69; 95% CI: 0.70–4.11; p = 0.25; Fig. 3A). However, in the analysis pooling four studies [42, 89, 104, 107] comparing HER2E and luminal A subtypes, HER2E was significantly associated with pCR outcome (OR: 12.11; 95% CI: 4.41–33.26; p < 0.0001; Fig. 3B). Similarly, when comparing against luminal B and luminal, combined, HER2E was significantly associated with pCR outcome (OR: 5.92; 95% CI: 2.59–13.54; p < 0.0001; Fig. 3C and OR: 3.37; 95% CI: 1.66–6.84; p = 0.0008; Fig. 3D, respectively). Substantial heterogeneity was reported for two of the pooled analyses: (1) in studies comparing HER2E and luminal-like (Fig. 3A); and (2) in studies comparing HER2E and luminal, combined (Fig. 3D). Although all three and seven studies pooled in the two analyses were performed in the Chinese population, the primary outcome assessed in Li et al. [45] and Xu et al. [42] focused on the contribution of genetic mutations or long non-coding RNAs (lncRNAs) as a predictor of pCR status in the recruited population. Consequently, pooled analysis excluding Li et al. eliminates the heterogeneity in the first analysis revealing significant association (OR: 2.30; 95% CI: 1.66–3.19; p < 0.00001), while pooled analysis excluding Li et al. and Xu et al. in the second analysis decreased the heterogeneity, and maintaining the significant association (OR: 3.80; 95% CI: 2.02–7.13; p < 0.0001).
Notably, when HER2E and TNBC patients were compared, neither was associated with pCR (OR: 1.46; 95% CI: 0.63–3.37; p = 0.38; Supplementary Fig. 1C, Additional File 7). We then compared luminal B and luminal A patients and found that luminal B patients were significantly associated with better pCR outcomes (OR: 3.26; 95% CI: 1.14–9.26; p = 0.03; Fig. 3E). Similar to NAC TA-treated patients, analyses conducted comparing the effect of TNBC and luminal subtypes on pCR outcome in TP-treated patients also significantly associate TNBC with better pCR outcome compared to luminal A (OR: 7.14; 95% CI: 2.82–18.04; p < 0.0001; Fig. 3F), luminal B (OR: 2.19; 95% CI: 1.09–4.41; p = 0.03; Fig. 3G), and luminal, combined (OR: 3.79; 95% CI: 1.94–7.40; p < 0.0001; Fig. 4A).
Our study compared the effect of chemotherapeutic agents on the pCR outcome in TNBC patients (Fig. 4B). Pooled analysis involving two studies [41, 89] showed that TNBC patients were significantly more likely to achieve pCR when treated with NAC TP (n = 25/62) compared to NAC TA (n = 7/48) (OR: 3.76; 95% CI: 1.43–9.87; p = 0.007).
Biomarkers
Qualitatively, most studies with biomarkers data comprise of routinely analysed biomarkers - ER, PR, HER2, and Ki-67. In ER- and ER + patients, anthracycline-based chemotherapy showed pCR rate of 27.2% (n = 4) vs 15.3% (n = 4). Meanwhile, TP, TA, and TAP chemotherapy showed pCR rate of 48.1% (n = 6) vs 19.4% (n = 5), 25.0% (n = 6) vs 9.0% (n = 7), and 29.7% (n = 1) vs 12.9% (n = 1), respectively. As for PR- and PR + patients, anthracycline-based, TP, TA, and TAP chemotherapy showed pCR rate of 28.3% (n = 2) vs 13.9% (n = 2), 46.6% (n = 5) vs 21.8% (n = 5), 20.6% (n = 4) vs 8.2% (n = 4), and 25% (n = 1) vs 13.3% (n = 1), respectively. Collectively, ER- and PR- breast cancer patients were likely to benefit more from TP regimen than TA, TAP, and anthracycline-based chemotherapy. Notably, some studies combined their report of ER and PR as Hormone Receptors (HR). Analysis of HR + and HR- patients showed pCR rate of 11.9% (n = 1) vs 10% (n = 1), 33.1% (n = 2) vs 49.3% (n = 2), 31.3% (n = 2) vs 50.7% (n = 3), and 12.4% (n = 5) vs 23.2% (n = 5) when treated with anthracycline-based, taxane-based, TP, and TA chemotherapy, respectively. Our findings suggested that HR + breast cancer patients achieved a better pCR rate when treated with a single-based chemotherapeutic agent, while HR- patients benefit more combination chemotherapy regimens.
In HER2 + and HER2- patients, anthracycline-based chemotherapy showed pCR rate of 10.5% (n = 3) vs 19.3% (n = 4). Meanwhile, TP, TA and TAP chemotherapy showed pCR rate of 44.6% (n = 15) vs 24.3% (n = 6), 20.1% (n = 14) vs 11.3% (n = 12) and 33% (n = 1) vs 13.5% (n = 1), respectively. One study [35] utilising CDK4/6 inhibitor on ER+/HER2- breast cancer patients and another study [119] utilising kinase inhibitor on ER+/HER2 + patients showed pCR rate of 5% (n = 1) and 0% (n = 1), respectively. Collectively, with the exception of anthracycline-based and targeted therapy, HER2 + breast cancer patients were likely to benefit more from TP, TA, and TAP regimens. Meanwhile, in patients with high and low Ki-67, anthracycline-based and taxane-based chemotherapy showed pCR rate of 12.2% (n = 1) vs 11.4% (n = 1) and 57.1% (n = 1) vs 32.1% (n = 1). When treated with TP and TA chemotherapy, the patients showed pCR rate of 39.9% (n = 5) vs 24% (n = 5) and 19.9% (n = 12) vs 7.7% (n = 12), respectively. Overall, breast cancer patients with high Ki-67 were likely to benefit more from taxane-based chemotherapy than TP, TA, and anthracycline-based regimen.
On the other hand, fourteen non-conventional biomarkers investigated in a few of the included studies were evaluated qualitatively. Three biomarkers – Bcl-2, Smac, and Survivin – were included for the evaluation of anthracycline-based chemotherapy. Anthracycline-based chemotherapy showed pCR benefit of 26.1% (n = 1) vs 4.3% (n = 1) in Bcl-2- and Bcl2 + breast cancer patients, 35.0% (n = 1) vs 8.6% (n = 1) in high and low Smac, and 28.3% (n = 1) vs 11.5% (n = 1) in low and high Survivin. Only one biomarker – ZEB1 - was evaluated for TP chemotherapy which revealed pCR rates of 36.1% (n = 1) in patients with low ZEB1 and 12.8% (n = 1) in patients with high ZEB1.
Ten biomarkers – Tau, P-gp, Topo-II, T-cadherin, CK5/6, EGFR, p53, LAG-3, cyclin D1, and nm23-H1 – were included for the evaluation of TA chemotherapy. Our findings showed pCR benefit of 31.3% (n = 1) vs 4.5% (n = 1) in Tau- and Tau+, 43.2% (n = 1) vs 7.7% (n = 1) in P-gp- and P-gp+, and 17% (n = 1) vs 3.4% (n = 1) in Topo-II- and Topo-II + breast cancer patients. The pCR rates observed in T-cadherin- and T-cadherin+, CK5/6- and CK5/6+, EGFR- and EGFR+, and p53- and p53 + breast cancer patients were 45.2% (n = 1) vs 7.4% (n = 1), 40% (n = 2) vs 25.6% (n = 2), 45.5% (n = 1) vs 28.1% (n = 1), and 33.3% (n = 1) vs 27.3% (n = 1), respectively. As for breast cancer patients with low and high expression of LAG-3, the pCR rates was observed at 64.7% (n = 2) vs 35.3% (n = 2). Lastly, breast cancer patients with cyclin D1 + and nm23-H1 + reported pCR benefits of 45.8% (n = 1) and 29.5% (n = 2) than 29.4% (n = 1) in cyclin D1- and 5.3% (n = 2) in nm23-H1- breast cancer patients.
Under meta-analysis, the role of biomarkers was investigated in NAC anthracycline-based and taxane-based -treated (Fig. 4), TA-treated (Figs. 4–6 and Supplementary Fig. 1, Additional File 7), and TP-treated (Figs. 6–7) breast cancer patients.
Anthracycline-based and taxane-based chemotherapy
Four studies [50, 62, 106, 125] were pooled for the effect of ER on pCR outcome in anthracycline-treated patients, where ER was not associated with pCR (OR: 1.95; 95% CI: 0.98–3.89; p = 0.06; Fig. 4C). Substantial heterogeneity was reported for the pooled analysis of ER- vs ER+, which could be explained by the study by Mohammadianpanah et al. [125] which was conducted in the Iranian population. In contrast, the other studies [50, 62, 106] were performed in the Chinese population. Subgroup analysis of ER- vs ER + pooling studies in the Chinese population only showed that ER- patients were significantly associated with pCR (OR: 2.52; 95% CI: 1.43–4.44; p = 0.001; Fig. 4C), supporting our hypothesis that the observed heterogeneity could be due to the difference in the Asian population. Notably, the observed moderate heterogeneity in the subgroup analysis can allude to by the differences in the population sizes of the three studies. Only two studies [50, 62] were pooled to analyse PR effect on pCR outcome in anthracycline-treated patients. Our findings showed that PR- patients were significantly associated with pCR (OR: 2.40; 95% CI: 1.52–3.80; p = 0.0002; Fig. 4D). Meanwhile, an analysis of three pooled studies [50, 62, 109] on the effect of HER2 on pCR outcome in anthracycline-treated breast cancer patients revealed that patients with HER2- biomarker were significantly more likely to achieve pCR (OR: 2.31; 95% CI: 1.42–3.75; p = 0.0008; Fig. 4E). Only one biomarker, HR, was evaluated for its effect on pCR outcome in Asian breast cancer patients treated with taxane-based chemotherapy (Fig. 4F). Two studies [43, 56] were pooled where patients with HR- biomarkers were significantly more likely to achieve pCR than HR- patients (OR: 1.96; 95% CI: 1.24–3.08; p = 0.004).
Taxane-anthracycline (TA) chemotherapy
Eight biomarkers comprising ER, PR, HR, HER2, nm23-H1, CK5/6, EGFR, and Ki-67 were investigated for their effect on pCR outcome in Asian breast cancer patients treated with NAC TA. Six studies [66, 73, 74, 81, 95, 112] and four studies [68, 69, 92, 110] were pooled to evaluate the association of ER and PR, respectively. Both ER- (OR: 3.19; 95% CI: 2.15–4.75; p < 0.0001; Fig. 4G) and PR- (OR: 3.11; 95% CI: 2.12–4.56; p < 0.0001; Fig. 5A) were significantly associated with pCR outcome in TA treated patients. Moderate heterogeneity was reported for the pooled analysis of ER- vs ER+. However, considering all the studies pooled for the analysis were performed in the Chinese population, the heterogeneity result was rejected.
In an analysis pooling five studies [48, 51, 91, 112, 128], HR was not associated with pCR outcome (OR: 2.38; 95% CI: 0.87–6.53; p = 0.09; Fig. 5B). The observed substantial heterogeneity could be due to the study by Elnemr et al. [128] conducted in Saudi Arabia, while the other four studies were performed in China. Consequently, subgroup analysis of HR- vs HR + removing the study by Elnemr et al. showed a decrease in the heterogeneity and significantly associated HR- with better pCR outcome when treated with TA (OR: 3.58; 95% CI: 1.62–7.90; p = 0.002). An analysis of nine pooled studies [51, 68, 69, 91, 92, 108, 110, 112, 128] showed that HER2 + is significantly associated with pCR (OR: 1.78; 95% CI: 1.05–3.02; p = 0.03; Fig. 5C) with substantial heterogeneity observed between the studies. Although the analysis also includes the study by Elnemr et al., which was conducted in Saudi Arabia, the heterogeneity could be influenced by the results pooled from seven studies that heavily pushed the effect of our analysis in one direction.
Our study synthesised meta-analysis data for other biomarkers apart from the commonly reported ones – ER, PR, HR, and HER2. In particular, two studies were evaluated for pCR outcome for nm23-H1 [12, 91] and CK5/6 [12, 105]. It was observed that nm23-H1- (OR: 6.74; 95% CI: 2.13–21.30; p = 0.001; Fig. 5D) and CK5/6- (OR: 1.87; 95% CI: 1.03–3.39; p = 0.04; Fig. 5E) are significantly associated with pCR. Two studies [12, 105] were pooled for analysis in the evaluation of pCR outcome with EGFR. Considerable heterogeneity was observed between the studies, perhaps alluded to the clinical differences between the studies of Li et al. [12] (n = 22/41) and Wang et al. [105] (n = 170/195) according to the distribution of patients with EGFR+. Despite that, EGFR is not associated with pCR outcome in TA-treated patients (OR: 2.02; 95% CI: 0.28-28.00; p = 0.38; Supplementary Fig. 1D, Additional File 7).
The proliferation index biomarker, Ki-67, was evaluated through an analysis of 12 pooled studies [12, 37, 48, 51, 61, 68, 69, 80, 91, 92, 105, 128], revealing significant association between pCR outcome and high Ki-67 (OR: 2.98; 95% CI: 1.79–4.97; p < 0.0001; Fig. 6A). The observed moderate heterogeneity between the studies could be due to differences in the Ki-67 cut-off value. Subgroup analysis pooling four studies [37, 69, 80, 128] with 14% Ki-67 cut-off did not significantly associate Ki-67 with pCR outcome (OR: 1.82; 95% CI: 0.65–5.10; p = 0.26). Significant heterogeneity was observed which could be explained by the primary research question addressed in the studies where Zhang et al. [80] focused on the prognostic value of magnetic resonance imaging (MRI), P-gp, and Ki-67, while the other three studies focused on the correlation of Ki-67 expression and pCR. Furthermore, the results pooled from the three studies heavily pushed the effect of our analysis in one direction. Thus, a pooled analysis excluding the study by Zhang et al. reveals null heterogeneity between the studies and a significant association between pCR and high Ki-67 with 14% cut-off value (OR: 3.12; 95% CI: 1.93–5.04; p < 0.00001). Meanwhile, subgroup analysis pooling five studies [48, 51, 61, 91, 105] with 20% Ki-67 cut-off significantly associate pCR outcome with high Ki-67 (OR: 2.88; 95% CI: 1.36–6.10; p = 0.006).
Taxane-platinum (TP) chemotherapy
Our study investigated five biomarkers comprising ER, PR, HR, HER2, and Ki-67 on their effect on pCR outcome in Asian breast cancer patients treated with NAC TP. Five studies [42, 64, 97, 104, 107] and two studies [100, 102] were pooled to evaluate the association of ER and PR, and HR, respectively. pCR outcome was significantly associated with ER- (OR: 4.91; 95% CI: 3.20–7.53; p < 0.00001; Fig. 6B), PR- (OR: 3.82; 95% CI: 2.48–5.90; p < 0.00001; Fig. 6C), and HR- (OR: 2.71; 95% CI: 1.43–5.15; p = 0.002; Fig. 6D).
Ki-67 was evaluated by analysing five pooled studies [42, 45, 64, 100, 107], showing that pCR outcome was not significantly associated with either high Ki-67 or low Ki-67 (OR: 2.20; 95% CI: 0.74–6.59; p = 0.16; Fig. 6E). The observed considerable heterogeneity was perhaps due to differences in the Ki-67 cut-off value, and four of the five studies heavily pushed the effect of our analysis to one direction. Subgroup analysis pooling two studies [42, 64] with 20% Ki-67 cut-off indicates that pCR outcome is significantly associated with high Ki-67 (OR: 4.37; 95% CI: 1.62–11.75; p = 0.003). This implicates the importance of having a standardised cut-off value for Ki-67, as at different cut-offs of 15% and 30%, neither Ki-67 biomarker was favoured as opposed to the 20% cut-off favours High Ki-67 to achieve pCR.
In the analysis of the effect of HER2 in breast cancer patients pooling five studies [42, 64, 97, 104, 107], pCR outcome was not significantly associated with neither HER2 + nor HER2- (OR: 2.44; 95% CI: 0.84–7.06; p = 0.10; Fig. 7A). The observed substantial heterogeneity could be influenced by the results pooled from four of the five studies that heavily pushed the effect of our analysis in one direction. Consequently, a pooled analysis excluding the study by Zhou et al. [64] reveals null heterogeneity between the studies and a significant association between pCR and HER2+ (OR: 4.14; 95% CI: 2.51–6.85; p < 0.00001).
Our study also evaluated the effect of chemotherapeutic agents on the pCR outcome in Asian breast cancer patients with HER2 + biomarker (Fig. 7B). Four studies [47, 58, 88, 99] were pooled and estimated. Our findings revealed neither NAC TP nor NAC TA was associated with pCR outcome in patients with HER2 + biomarker (OR: 1.60; 95% CI: 0.66–7.06; p = 0.30). Substantial heterogeneity was reported which could be explained by the difference in the study design of the pooled studies. Of the four studies, Huang et al. [99] was the only study that conducted a randomised controlled trial (RCT) where the recruited HER2 + breast cancer patients were assigned to either TA or TP chemotherapy by the investigator. In contrast, HER2 + breast cancer patients recruited in the other three studies [47, 58, 88] were given either TA or TP regimen based on their preferences. Notably, subgroup analysis excluding Huang et al. decreased the heterogeneity I2 and significantly associated HER2 + breast cancer patients with better pCR outcome when treated with TP (OR: 2.36; 95% CI: 1.07–5.20; p = 0.03).
Genetics variations and differential expression
Eleven genes – PIK3CA, TP53, EPIC1, TOP2A, ERBB2, MYC, CCND1, PCDH17, EPIC1, BRCA1 and BRCA2 – were included for the qualitative evaluation of NAC regimens. No specific single variant vs wildtype was compared for most of the genes since most of the evaluated studies did not report them. Therefore, our analysis only compared wildtype (wt) and mutated (mt), where the mutated gene might contain single or multiple variants.
Breast cancer patient harbouring wt and mutated mt PIK3CA showed pCR rate of 19.4% (n = 1) vs 14.1% (n = 1) and 18.8% (n = 1) vs 16.1% (n = 1) when treated with anthracycline-based and taxane-based chemotherapy, respectively. Meanwhile, patients treated with TA chemotherapy showed a pCR rate of 21.3% (n = 2) vs 9.4% (n = 2). Thus, breast cancer patients with wt PIK3CA were likely to benefit more from TA regimen than anthracycline-based and taxane-based chemotherapy. Interestingly, breast cancer patient harbouring wt and mt TP53 showed pCR rate of 7.1% (n = 1) vs 28.6% (n = 1), 11.3% (n = 1) vs 15.2% (n = 1), and 6.1% (n = 1) vs 16.1% (n = 1) when treated with anthracycline-based, taxane-based, and TA chemotherapy, respectively. Our findings suggested that breast cancer patients with mt TP53 were likely to benefit more from anthracycline-based chemotherapy than taxane-based and TA regimen.
Our findings also showed that breast cancer patients with TOP2A, ERBB2, and MYC amplification (amp) achieved higher pCR rates than wt TOP2A, ERBB2, and MYC (56.3% (n = 1) vs 13.8% (n = 1), 28.4% (n = 1) vs 6.1% (n = 1), and 13.7% (n = 1) vs 11.2% (n = 1), respectively) when treated with TA regimens.On another note, breast cancer patients with wt CCND1 and unmethylated (unm) PCDH17 achieved higher pCR rate than those with CCND1 amp andmethylated (m) PCDH17 (13.8% (n = 1) vs 2.7% (n = 1) and 67.3% (n = 1) vs 31.6% (n = 1), respectively).
One included study by Mou et al. [60] focused on the effect of UGT2B7 rs7435335 on NAC TA efficacy. It was observed that patients with the genotype GA achieved higher pCR rate (42.3% (n = 1)) than patients with the genotype GG (18.9% (n = 1)). Another study by Xu et al. [94], analysed the effect of BRCA1 and BRCA2 mRNA expression in breast cancer patients treated with anthracycline-based and taxane-based chemotherapy. Our findings showed pCR benefit of 24.6% (n = 1) vs 16.9% (n = 1), 16.9% (n = 1) vs 17.5% (n = 1), and 14% (n = 1) vs 20.8% (n = 1) in anthracycline-based treated patients with low, intermediate, and high BRCA1 vs BRCA2 mRNA expression, respectively. Meanwhile, in taxane-based treated patients, our findings showed pCR benefit of 19.6% (n = 1) vs 24.4% (n = 1), 26.8% (n = 1) vs 23.4% (n = 1), and 21.4% (n = 1) vs 18.9% (n = 1) with low, intermediate, and high BRCA1 vs BRCA2 mRNA expression, respectively. Notably, breast cancer patients with low EPIC1 showed higher pCR rate (40.7% (n = 1)) when treated with TP regimen than patients with high EPIC1 (33.3% (n = 1)).
Under meta-analysis, only one gene was analysed for its effect on pCR outcome in Asian breast cancer patients treated with NAC TA (Fig. 7).
Taxane-anthracycline (TA) chemotherapy
From the analysis of two studies [51, 79], 564 patients were pooled for PIK3CA analysis. It was observed that patients harbouring wt PIK3CA were significantly associated with better pCR outcomes compared to patients with mt PIK3CA gene (OR: 2.44; 95% CI: 1.42–4.19; p = 0.001; Fig. 7).
Pooled reported association
Meta-analyses of pooled reported association of pCR were evaluated according to molecular classification, genetic variations, and biomarkers characterisation of the Asian breast cancer patients.
Molecular classification
An adjusted pooled analysis of TNBC against non-TNBC patients showed that TNBC patients were significantly associated with better response when treated with neoadjuvant chemotherapy (OR: 3.02; 95% CI: 1.54–5.95; p = 0.001; Supplementary Fig. 2A, Additional File 7). The observed moderate heterogeneity could be due to the study by Lv et al. [89] that did not specifically study TNBC vs non-TNBC patients. Moreover, the recruited patients were either treated with anthracycline-based or TP or TA regimens. Meanwhile, Wu et al. [53] specifically recruited TNBC and non-TNBC patients, and all were treated with TA. Hence, Wu et al.’s result carried more weight than Lv et al. in this analysis. However, Lv et al. compensated for these differences by adjusting their multivariate analysis with molecular subtypes. Overall, this result should be taken with caution.
Genetic variations
Amongst pooled reported associations of pCR for genetic variations, our study evaluated the effect of PIK3CA and TP53 genes in NAC-treated Asian BC patients. Notably, no specific or single variant vs wildtype was addressed for both genes as well. In the adjusted analysis of three [51, 79, 99] and two studies [51, 99] for PIK3CA and TP53 genes, respectively, breast cancer patients harbouring mutation in the PIK3CA gene was associated with worse response (OR: 0.64; 95% CI: 0.42–0.98; p = 0.04; Supplementary Fig. 2B, Additional File 7) while TP53 gene was not associated with pCR outcome (OR: 1.34; 95% CI: 0.59–3.05; p = 0.49; Supplementary Fig. 2B, Additional File 7).
Biomarkers
Amongst pooled reported associations of pCR for biomarkers, our study evaluated the effect of Tau, nm23-H1, ER, PR, HR, HER2, and Ki-67 biomarkers in NAC-treated Asian BC patients (Supplementary Figs. 2–8, Additional File 7). In an adjusted analysis of Tau pooling two studies [76, 87], the result suggests that Tau + was associated with worse response in the neoadjuvant setting (OR: 0.22, 95% CI: 0.09–0.54, p = 0.0008; Supplementary Fig. 2C, Additional File 7). Meanwhile in adjusted analysis of nm23-H1 pooling two studies [12, 91], nm23-H1 was not associated with pCR outcome in TA-treated patients (OR: 1.56, 95% CI: 0.55–4.45, p = 0.41; Supplementary Fig. 2C, Additional File 7).
The association between pCR in NAC-treated breast cancer patients and ER was evaluated in Supplementary Fig. 3 (see Additional File 7). In the crude analysis of ER + vs ER- pooling five studies [42, 54, 73, 76, 107], ER was not associated with pCR outcome (OR: 0.39; 95% CI: 0.13–1.15; p = 0.09). The reported moderate heterogeneity between the studies was perhaps due to the difference in favoured outcomes in one study [73] compared to the rest. Moreover, the heterogeneity is attributable to the different NAC treatments received in each study, whereby the patients were either treated with TP or anthracycline-containing regimens. Therefore, subgroup analyses of ER- vs ER + pooling studies with patients treated with anthracycline-containing chemotherapy [73, 76] and taxane-platinum chemotherapy [42, 54, 107] were conducted. Our analysis revealed that ER was not associated with pCR outcome when patients were treated with an anthracycline-containing agent (OR: 1.19; 95% CI: 0.07–19.28; p = 0.90) with considerable heterogeneity observed between the studies. The heterogeneity is perhaps due to differences in the favoured outcome in each study caused by the addition of the taxane regimen with the anthracycline in Li et al. [76]. Contrarily, ER + patients were significantly associated with worse response when treated with TP (OR: 0.19; 95% CI: 0.11–0.32; p < 0.00001).
Meanwhile, adjusted analysis of ER + vs ER- pooling fourteen studies [42, 44, 50, 54, 55, 60, 70, 73, 76, 79, 84, 87, 92, 132] also showed that ER was not associated with pCR outcome (OR: 0.59; 95% CI: 0.32–1.08; p = 0.09; Supplementary Fig. 3, Additional File 7). The observed substantial heterogeneity between the studies was perhaps due to differences in the chemotherapeutic agents received in each study. Subgroup analysis pooling five studies [44, 55, 79, 87, 132] with patients treated in the neoadjuvant setting revealed that ER was not associated with pCR outcome (OR: 0.47; 95% CI: 0.19–1.14; p = 0.09) with considerable heterogeneity reported between the studies probably due to clinical variances between the studies. Another two subgroup analyses pooling studies with patients treated with TP [42, 54] and TA [60, 76, 92] regimens indicate that ER + was significantly associated with worse response (OR: 0.21; 95% CI: 0.06–0.70; p = 0.01 and OR: 0.34; 95% CI: 0.19–0.61; p = 0.0003, respectively). While subgroup analysis pooling studies with patients treated with anthracycline-based chemotherapy [50, 73] showed that ER was not associated with pCR response (OR: 4.29; 95% CI: 0.67–27.39; p = 0.12). Substantial heterogeneity was reported probably due to the difference in effect size between the studies attributed to the study sample size, whereby Zhao et al. [73] recruited 98 locally advanced breast cancer patients while Yao et al. [50] recruited 538 breast cancer patients.
Meanwhile, analyses pooling studies based on the characteristics of the patients showed that ER was not associated with pCR in patients achieving a complete pathological response in the breast only (OR: 0.93; 95% CI: 0.18–4.85; p = 0.94; Supplementary Fig. 3, Additional File 7). The observed heterogeneity between the studies is probably due to the smaller number of studies pooled (n = 2) and each study favoured a different outcome. Notably, ER- patients were significantly associated with pCR in studies pooling anthracycline-treated patients (OR: 2.78; 95% CI: 1.61–4.78; p = 0.0002; Supplementary Fig. 3, Additional File 7).
The role of the biomarker PR was assessed in Asian breast cancer patients subjected to neoadjuvant chemotherapy (Supplementary Fig. 4, Additional File 7). In an analysis with crude OR results pooling five studies [42, 54, 73, 76, 107], it was observed that patients with PR + were significantly associated with worse response (OR: 0.40; 95% CI: 0.20–0.79; p = 0.009). The observed substantial heterogeneity between the studies was perhaps due to the difference in the favoured outcome in one study [73] and differences in weightage and population size. Subgroup analysis pooling studies with TP-treated patients [42, 54, 107] revealed PR + patients were significantly associated with worse response (OR: 0.29; 95% CI: 0.13–0.62; p = 0.001). Although moderate heterogeneity was reported, all the three studies pooled for the subgroup analysis were conducted in the Chinese population with no clinical variances. Thus, the heterogeneity result was rejected. Meanwhile, in subgroup analysis pooling patients treated with anthracycline-containing NAC [73, 76], the biomarker PR was not associated with treatment response (OR: 0.66; 95% CI: 0.17–2.62; p = 0.55).
PR was also not associated with treatment response in the adjusted analysis pooling eight studies [42, 44, 50, 54, 76, 79, 84, 92] (OR: 1.01; 95% CI: 0.64–1.60; p = 0.97; Supplementary Fig. 4, Additional File 7). The observed moderate heterogeneity was probably due to differences in population size and differences in treatment given in each study. We then conducted four subgroup analyses focusing on the treatment regimen and found that PR was not associated with treatment response in breast cancer patients treated: (1) in the neoadjuvant setting (OR: 1.01; 95% CI: 0.47–2.18; p = 0.99); (2) with TP (OR: 0.94; 95% CI: 0.34–2.58; p = 0.91); and (3) TA regimens (OR: 0.37; 95% CI: 0.05–2.94; p = 0.35), but PR- was significantly associated with better response in anthracycline-containing treated patients (OR: 1.63; 95% CI: 1.03–2.57; p = 0.04). Meanwhile, in adjusted analysis pooling studies with anthracycline-treated patients [62, 94] revealed that PR was not significantly associated with treatment response (OR: 1.43; 95% CI: 0.74–2.77; p = 0.29).
The association between treatment response and hormone receptors (HR) comprising ER and PR was assessed in Asian breast cancer patients (Supplementary Fig. 5, Additional File 7). In an analysis with crude OR pooling four studies [63, 99, 113, 118], it was observed that patients with HR + were significantly associated with worse treatment response in the neoadjuvant setting (OR: 0.47; 95% CI: 0.24–0.92; p = 0.03). In a subgroup analysis whereby all the recruited patients in the pooled studies [63, 99] were of HER2+, it was observed that HR + were significantly associated with a worse response (OR: 0.40; 95% CI: 0.18–0.89; p = 0.02). Excluding the aforementioned studies, subgroup analysis pooling two studies [113, 118] revealed that HR was not significantly associated with treatment response in the neoadjuvant setting (OR: 1.16; 95% CI: 0.22–6.22; p = 0.87).
Similarly, analysis with adjusted OR pooling five studies [63, 91, 99, 113, 116] indicated that HR was not significantly associated with treatment response (OR: 1.27; 95% CI: 0.47–3.45; p = 0.64; Supplementary Fig. 5, Additional File 6). The observed substantial heterogeneity was perhaps due to the clinical variance in the characteristics of the recruited breast cancer population in each study. Subgroup analysis pooling studies analysing HR + vs HR- in the neoadjuvant setting [113, 116] showed that HR- were significantly associated with better treatment responses (OR: 2.39; 95% CI: 1.17–04.87; p = 0.02). Meanwhile, subgroup analysis pooling studies with HER2 + patients [63, 99] revealed that HR+/HER2 + breast cancer were significantly associated with worse treatment responses in the neoadjuvant setting (OR: 0.43; 95% CI: 0.21–0.88; p = 0.02).
The association between pCR outcome and HER2 was estimated in Supplementary Fig. 6 (see Additional File 7). In an analysis pooling crude OR of seven studies [42, 54, 73, 76, 81, 107, 113], it was observed that HER2 + breast cancer patients were significantly associated with better treatment response (OR: 2.50; 95% CI: 1.44–4.35; p = 0.001). The moderate heterogeneity reported between the studies was perhaps due to the differences in treatment regimens given to the recruited breast cancer population in each study. Subsequently, two subgroup analyses pooling studies according to the chemotherapy regimens administered to the breast cancer patients revealed that HER2 + patients treated with TP regimen [42, 54, 107] were significantly associated with better response (OR: 4.64; 95% CI: 2.74–7.86; p < 0.00001), while HER2 was not associated with treatment response in patients treated with anthracycline-containing regimen (84,141) (OR: 2.08; 95% CI: 0.90–4.78; p = 0.09).
Similarly, in an analysis pooling adjusted OR of 12 studies [42, 44, 50, 54, 76, 79, 81, 84, 87, 91, 116, 132], it was observed that HER2 + breast cancer patients were significantly associated with better treatment response (OR: 2.29; 95% CI: 1.56–3.35; p < 0.0001; Supplementary Fig. 6, Additional File 7). Substantial heterogeneity was reported, perhaps due to the clinical variances in each study, based on the treatment received by the patients and the characteristics of the recruited breast cancer population. Subgroup analysis pooling five studies in the neoadjuvant setting [44, 79, 87, 116, 132] showed patients with HER2 + were significantly associated with better response (OR: 2.33; 95% CI: 1.31–4.15; p = 0.004). Some differences between the studies might explained the observed substantial heterogeneity: (1) two of the studies [44, 87] main objective was to investigate the association of Tau with response to neoadjuvant chemotherapy, while Yuan et al. [79] focused on the association of PIK3CA mutation status with response to neoadjuvant chemotherapy, and Lim et al. [132] and Lv et al. [116] assessed factors affecting neoadjuvant treatment response; (2) Lim et al. [132] was the only study including multi-ethnic cohort of breast cancer patients since it was conducted in Singapore and Malaysia, although one of the ethnicity included in Lim et al. was Chinese. Meanwhile, subgroup analyses pooling two studies of patients treated with TP [42, 54] and two studies of patients treated with anthracycline-containing chemotherapy [50, 84] indicate HER2 + patients were significantly associated with better treatment response (OR: 7.07; 95% CI: 2.88–17.40; p < 0.0001 and OR: 2.65; 95% CI: 1.66–4.23; p < 0.0001, respectively). However, subgroup analysis pooling two studies of patients treated with TA [76, 91] revealed that HER2 was not associated with treatment response (OR: 1.08; 95% CI: 0.47–2.51; p = 0.85).
Notably, HER2 was also not associated with pCR in patients achieving a complete pathological response in the breast only (OR: 1.96; 95% CI: 0.78–4.89; p = 0.15; Supplementary Fig. 6, Additional File 7) and in anthracycline treated patients (OR: 1.52; 95% CI: 0.97–2.40; p = 0.07; Supplementary Fig. 6, Additional File 7).
As for Ki-67, an analysis pooling crude OR of eight studies [42, 54, 63, 65, 76, 99, 107, 124] observed that patients with high Ki-67 were significantly associated with better treatment responses (OR: 2.63; 95% CI: 1.69–4.07; p < 0.0001; Supplementary Fig. 7, Additional File 7). Subgroup crude analyses pooling studies with TNBC patients [65, 124] and patients treated with TP [42, 54, 107] and TA [65, 76] revealed that patients with high Ki-67 were significantly favoured to achieve better treatment responses (OR: 4.42; 95% CI: 1.41–13.85; p = 0.01, OR: 2.13; 95% CI: 1.21–3.75; p = 0.009, and OR: 4.26; 95% CI: 1.90–9.54; p = 0.0004, respectively). Meanwhile, subgroup crude analysis pooling studies with HER2 + patients [63, 99] showed that Ki-67 was not associated with treatment response (OR: 1.55; 95% CI: 0.79–3.07; p = 0.20).
Similarly, an analysis of adjusted OR pooling seven studies [12, 42, 54, 65, 88, 92, 99] showed that patients with high Ki-67 were significantly associated with better treatment response (OR: 2.63; 95% CI: 1.56–4.41; p = 0.0003; Supplementary Fig. 7, Additional File 7). In subgroup adjusted analysis pooling studies with HER2 + patients [88, 99], breast cancer patients with high Ki-67 were significantly associated with better response in the neoadjuvant setting (OR: 3.67; 95% CI: 1.11–12.12; p = 0.03). Congruent with the subgroup crude analysis, a subgroup adjusted analysis of pooled studies with TNBC patients [12, 65] also revealed patients with high Ki-67 were significantly associated with better treatment response (OR: 2.16; 95% CI: 1.00-4.64; p = 0.05). Notably, the analysis was also influenced by the fact that the TNBC patients were treated with the TA regimen. Breast cancer patients with high Ki-67 were also significantly associated with better treatment response when treated with TA regimen (OR: 2.24; 95% CI: 1.34–3.74; p = 0.002). However, in patients treated with NAC TP, Ki-67 was not associated with treatment response (OR: 4.63; 95% CI: 0.35–61.14; p = 0.24).
Three biomarkers - ER, HR, and Ki-67 - were evaluated by pooling studies reporting their association using hazards ratio (Supplementary Fig. 8, Additional File 7). Our pooled adjusted analysis of two studies [40, 68] showed that ER- patients were significantly associated with better treatment response (HR: 2.75; 95% CI: 1.25–6.05; p = 0.01). Both crude and adjusted analysis of HR- vs HR + and high vs low Ki-67 in HER2 + patients treated with taxane-containing chemotherapy showed that HR- patients and patients with high Ki-67 were significantly associated with a better response. However, in an adjusted result analysis of high vs low Ki-67 in patients treated with taxane-containing chemotherapy, Ki-67 was not significantly associated with treatment response (HR: 1.26; 95% CI: 1.26–8.25; p = 0.81) with substantial heterogeneity. The significant heterogeneity might be explained by the variation in the taxane-based treatment regimen where Zhang et al. [43] included patients submitted to either single taxane-based regime or taxane-platinum combination, while all patients recruited in Wang et al. [68] and Ding et al. [100] were submitted to taxane-anthracycline and taxane-platinum combination regimens, respectively. Moreover, Zhang et al. and Ding et al. incorporated trastuzumab as part of their neoadjuvant regimen while Wang et al. subjected their patients to trastuzumab in the adjuvant setting.
Publication bias
Publication bias assessment was done using the Jamovi Software (version 2.3) [134] (see Supplementary Figs. 1–13, Additional File 8). The occurrence of publication bias was observed in two analyses.
First, in the overall analysis evaluating the association of pCR outcome with HER2 + and HER2- breast cancer patients submitted to TA chemotherapy, the regression test indicated funnel plot asymmetry (p = 0.03) but not the rank correlation test (p = 0.34). File drawer analysis indicated that at least 51 studies would be required to nullify the effect (p < 0.001). Hence, there is less chance of publication bias in the analysis. As indicated in Fig. 5, subgroup analysis was not conducted for the overall analysis because although the effect is estimated to favour HER2 + significantly, in some studies the true effect may in fact favour HER2-.
Second, in the overall analysis evaluating the association of treatment response with PR- and PR + breast cancer patients in the neoadjuvant setting, the rank correlation test indicated funnel plot asymmetry (p = 0.03) but not the regression test (p = 0.08). File drawer analysis suggested the presence of publication bias in the analysis, which could be explained by the heterogeneity observed between the pooled studies, specifically in the treatment regimens assessed in each study. Thus, four subsequent subgroup analyses – pooling two studies at each treatment regimen the patients were subjected to – were conducted for the overall PR- vs PR+ (adjusted results) analysis (Supplementary Fig. 4, Additional File 7).