Xpert^® Breast Cancer STRAT4 Assay Using Fine Needle Aspiration Biopsy Samples in a Resource Constrained Setting: A Prospective Diagnostic Accuracy Study

Abstract

Background

Use of fine needle aspiration biopsy (FNAB) specimens on Xpert^® Breast Cancer STRAT4 (CEIVD) could potentially increase access to breast cancer biomarker testing in resource-constrained settings. We assessed the performance of a Research Use Only (RUO) version of STRAT4 using FNAB specimens in Tanzania.

Methods

Patients ≥18 years with palpable breast masses presenting to the FNAB Clinic at Muhimbili National Hospital were recruited consecutively. Patients who were pregnant, lactating, or had a prior diagnosis of breast cancer were excluded. STRAT4 testing was performed on FNAB samples using four protocols: 1x (1x-FNAB), quick lysis (QL), and Maui (M) on FNAB smears, and 1x on cell block (1x-CB). For 1x-FNAB and 1x-CB, tissue was processed using FFPE lysis reagent, incubated at 80°C with proteinase K, and followed by ≥ 95% ethanol. QL was processed using FFPE lysis reagent and ≥ 95% ethanol, while M was processed using a proprietary RUO polyethylene glycol-based lysis reagent. The primary outcomes were overall concordance, sensitivity, specificity, and area under receiver operating characteristic curve (AUC) of STRAT4 as compared to immunohistochemistry and/or in situ hybridisation results performed on cell blocks using clinically validated protocols in a Clinical Laboratory Improvement Amendments-accredited laboratory.

Findings

Between November 29, 2017-December 17, 2020, 208 Tanzanian women were enrolled. For estrogen receptor, 1x-FNAB had the best performance with overall concordance rate of 95% [95% CI 90–100], sensitivity 94% [95% CI 85–100%], specificity 97% [95% CI 90–100%), and AUC 0·96. For progesterone receptor, 1x-FNAB performed the best among smear protocols with concordance of 84% [95% CI 74–93%], sensitivity 63% [95% CI 43–82%], specificity 97% [95% CI 92–100%], and AUC 0·91. For human epidermal growth factor receptor 2, M had highest agreement with overall concordance rate of 93% [95% CI 85–96%], sensitivity 96% [95% CI 88–100%], specificity 92% [95% CI 87–98%] and AUC of 0·94. For Ki-67, M had the best performance of smear protocols with concordance of 69% [95% CI 60–79%], sensitivity 62% [95% CI 50–74%], specificity 84% [95% CI 72–97%], and AUC 0·78.

Interpretation

Processing FNAB samples with STRAT4 is feasible in Tanzania, and performance for ER is robust. Further optimisation of STRAT4 for FNAB has potential to improve timely access to breast cancer diagnostics in resource-constrained settings.

INTRODUCTION

Pathologic diagnosis, including estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2) testing, are critical in determining optimal treatments for breast cancer. However, a recent survey of laboratories in sub-Saharan Africa (SSA) found that only half provide onsite immunohistochemistry (IHC) services.¹ Innovative solutions to improve availability of cancer biomarker testing in resource-constrained settings are needed.

Fine needle aspiration biopsy (FNAB) is often the primary mode of tissue diagnosis in SSA.² FNAB is low-cost and has high sensitivity (92–98%) and positive predictive values (92–100%) for malignant breast masses when performed by skilled operators.³ Several studies have established strong concordance of ER, PR, and HER2 expression between cell block (CB) and histologic specimens.^4–6 FNAB allows for rapid onsite evaluation (ROSE) to evaluate for specimen adequacy and immediate preliminary diagnosis.^7,8

The Xpert^® Breast Cancer STRAT4 assay (Cepheid, Sunnyvale, CA, USA) is a validated CE-marked in vitro diagnostic (CE-IVD) automated cartridge-based multiplex polymerase chain reaction assay that runs on the GeneXpert^® Instrument System (Cepheid, Sunnyvale, CA, USA) and delivers semi-quantitative messenger ribonucleic acid (mRNA) levels for ESR1, PGR, ERBB2 and MKI67 in FFPE breast cancer specimens.⁹ STRAT4 has been proposed as a solution to limitations in IHC in LMICs due to low demands for training and reagents. Moreover, it does not rely upon pathologist interpretation. The performance of STRAT4 has been shown to have high concordance with IHC when using formalin-fixed paraffin embedded tissue (FFPE).^9,10

Pairing STRAT4 with FNAB may render traditional histologic methods unnecessary, offering exciting opportunities for this technology to be implemented in facilities with limited pathology infrastructure to avail timely and accurate results for breast cancer patients. We aimed to assess the performance of a Research use only (RUO) version of STRAT4 to provide ER, PR, HER2, and Ki-67 results using off-label FNAB specimens in Tanzania.

METHODS

Study design and participants

This prospective study was conducted at Muhimbili National Hospital (MNH) in Tanzania. Tanzanian patients aged ≥18 years who presented to the FNAB Clinic in the MNH Central Pathology Laboratory with palpable breast masses were recruited using consecutive sampling. Individuals who were pregnant, lactating, or had a prior diagnosis of breast cancer were excluded. If bilateral breast masses were detected, only one site was selected and laterality noted. A research assistant abstracted clinical information from the medical record including age, sex, mass location, clinical tumor size, and the presence/absence of palpable lymphadenopathy. Data on race/ethnicity was not collected. We planned to recruit patients 200 patients, based on College of American Pathologists (CAP) guidelines for test validation of 40 positive and 40 negative challenges for each unique biomarker.^11,12

The study was approved by institutional review boards at the Tanzanian National Institute of Medical Research (NIMR/HQ/R.8a/Vol. IX/2783), Muhimbili University of Health and Allied Sciences (MUHAS) (DA.287/298/01.A/) and University of California, San Francisco (UCSF) (17–22963). Written informed consent was obtained from all participants.

The decision to use FNAB cell blocks was based on an analysis preceding study design. Access to core biopsy is limited in Tanzania, and procedures were not in place to ensure adherence to American Society of Clinical Oncology (ASCO)/CAP guidelines.^13,14 Several studies have established strong concordance of ER, PR, and HER2 expression between CBs and histologic specimens.^4,5

Specimen acquisition

Specimen acquisition occurred in the MNH FNAB Clinic as previously described (appendix p 1).⁶ FNAB and ROSE were performed by two trained pathologists (AHK and MCN). For participants with suspicious or malignant masses as determined by ROSE using toluidine blue stain, a trained pathologist performed 1–2 dedicated FNAB pass(es), and material was used to create air-dried direct smears for STRAT4 testing and/or rinsed into 10% neutral buffered formalin and fixed for 6–72 hours. Before completing the procedure, the air-dried direct smears and formalin vials were checked for the presence of tissue fragments.

Cell block processing

FFPE CB were processed in MNH Central Pathology Laboratory as previously described (appendix p 1).⁶ Hematoxylin and eosin-stained slides from CBs were reviewed for adequate cellularity. For ER and PR IHC, adequate cellularity was defined as the presence of ≥ 1 well-preserved cluster of 50 epithelial cells. For HER2 and Ki-67, adequate cellularity was defined as ≥ 1 well-preserved cluster of 100 epithelial cells. CB with insufficient epithelial cells were excluded.

Immunohistochemistry

For quality assurance, IHC tests were performed on CBs at UCSF and compared with IHC performed at MNH on corresponding specimens according to each institution’s standard operating procedures (appendix pp 1–3). At UCSF, IHC and FISH results were interpreted according to ASCO/CAP guidelines and while blinded to STRAT4 results.^13,14 A cut-point of 20% was used for Ki-67 IHC.¹⁵ At MNH, ER and PR IHC were interpreted using Allred score,¹⁶ and HER2 IHC using ASCO/CAP guidelines.¹³ Ki-67 IHC was not performed at MNH because of resource constraints.

HER2 fluorescence in situ hybridisation (FISH)

All equivocal (2+) cases by HER2 IHC were submitted for FISH at UCSF, using the PathVysion HER2 DNA Probe Kit (Abbott Laboratories, Abbott Park, IL). Interpretation was performed following ASCO/CAP guidelines (appendix p 3).¹³

STRAT4 Assay

Breast cancer specimens were tested with STRAT4 using at least one of four methods: “1x protocol” on FNAB smears (1x-FNAB), “1x protocol” on FFPE CB material (1x-CB), “Quick Lysis protocol” (QL) on FNAB smears, or “Maui protocol” (M) on FNAB smears (Figure 1). STRAT4 RUO mRNA assays were performed at the MUHAS Pathology Laboratory. GeneXpert DX software settings were set for the STRAT4 RUO assay as previously described,⁹ except for use of modified delta Ct (dCt) assay cutoff values. The modified dCt cutoffs used in the current study were previously identified in pooled analysis of >1000 FFPE breast cancer tissue specimens (unpublished data), which vary only slightly from the original STRAT4 dCt cutoffs. However, the modified cutoffs reflect optimal concordance for each STRAT4 marker versus IHC/FISH across a larger, real-world sampling of specimens and testing methods across different geographical regions. The modified dCt cutoffs for each marker used were: ESR1 positive, dCt ≥ −0·5; PGR positive, dCt ≥ −1·7;ERBB2 positive, dCt ≥−0·25; and MKi67 positive (high), dCt ≥ −2·6. For ESR1 and ERBB2, assays were interpreted as positive or negative, or invalid. For PGR and MKi67, assays were interpreted as positive or negative, indeterminate, or invalid. A valid negative result for PGR and MKi67 required a CYFIP1 reference gene Ct of ≤31. If the CYFIP1 Ct was >31, a “negative” result would be classified as „indeterminate.’ Final positive or negative results for each biomarker were then analysed for concordance against IHC and/or FISH results in CB samples. HER2-low status was not accounted for in selection of these cut-offs but is under investigation by the manufacturer. Given the automated design, it was not necessary to blind operators of STRAT4 to IHC/FISH results.

Figure 1.

Flow diagram of sample collection and testing. Abbreviations: FNAB, fine-needle aspiration biopsy; CB, cell block; ER, estrogen receptor; PR, progesterone receptor; HER2, human epidermal growth factor receptor 2; IHC, immunohistochemistry; FISH, fluorescence in situ hybridisation; UCSF, University of California, San Francisco; STRAT4, Xpert^® Breast Cancer STRAT4 Assay; 1x-FNAB, 1x protocol on fine needle aspiration biopsy smears; QL, quick lysis protocol on fine needle aspiration biopsy smears; M, Maui protocol on fine needle aspiration biopsy smears; 1x-CB, 1x protocol on cell block;

1x Protocols – “1x-FNAB” and “1x-CB”

1x protocol was performed as described previously.¹⁷ For 1x-FNAB, an unstained, air-dried FNAB smear was scraped off a glass slide using a razor blade directly into a 1·5ml tube. Alternatively, for CBs, a 4μm unstained section placed on a glass slide was scraped into a 1·5ml tube. Then 1·2ml of Xpert FFPE lysis reagent and 20 μl proteinase K was added. Samples were vortexed, pipette mixed until blood clots were lysed, and then incubated at 80°C for 30 minutes. Samples were transferred to a 5ml vial, and 1·2 ml ≥ 95% ethanol was added and vortexed again. A 520μl aliquot of the sample was then transferred to a STRAT4 cartridge for testing.

Quick Lysis and Maui Protocols

QL and M were introduced as simplified processing methods with reduced requirements for reagents and incubation time. QL was routinely performed starting at 17 months, and M was introduced at 10 months. In QL, material from an unstained, air-dried FNAB smear was scraped off a glass slide directly into a 5ml tube. Then 1·2ml of Xpert FFPE lysis reagent was added. Samples were vortexed, pipette mixed until blood clots were lysed. 1·2 ml of ≥ 95% ethanol was added, Next, samples were vortexed again for 15 seconds, then quick spun. In M, 1·2 ml of a proprietary RUO polyethylene glycol-based lysis reagent (i.e. “Maui reagent”) was added to tubes with FNAB material and the lysate was vortexed for 15 seconds. Finally, for both QL and M, a 520μl aliquot of lysate was transferred into a STRAT4 cartridge.

Outcomes

The main objective of the study was to determine overall concordance, sensitivity, specificity, and area under the curve (AUC) of receiver operating characteristic curves (ROC) of ER, PR, HER2, and Ki-67 for each STRAT4 protocol, as compared to the results of IHC and/or ISH performed on cell blocks at UCSF’s Clinical Laboratory Improvement Amendments-accredited laboratory.

Statistical analysis

STRAT4 and IHC/FISH results (positive vs. negative) were tested for agreement. Percent concordance with its 95% confidence interval (CI) was calculated. Cohen’s Kappa statistic with 95% CI was used to assess agreement, whereby a value of 0·01 to 0·20 indicated slight concordance, 0·21 to 0·40 fair, 0·41 to 0·60 moderate, 0·61 to 0·80 substantial, and 0·81 to 0·99 strong concordance. McNemar’s Chi-squared test was used to test that the ratio of discordant proportions as false positives to false negatives equals one. Referencing UCSF IHC/FISH results as standard, sensitivity, specificity, positive predictive value, and negative predictive value with 95% CI of STRAT4 results were calculated. If a case had STRAT4 or IHC/FISH data that was indeterminate, invalid, or missing for a particular biomarker, the case was excluded from analysis for that biomarker.

ROC curves were generated from logistic regression models that fit IHC results to corresponding STRAT4 quantitative values. Then, 5-fold cross-validation was used to obtain the AUC with 95% CI. Statistical significance was established at p < 0·05. All analyses were performed with Stata, version 18 (StataCorp. 2023) and SAS, version 9.4 (SAS Institute, Inc).

Role of funding source

The funders had no role in the data collection, data analysis, or data interpretation. Cepheid contributed to study design, editing of the methods section of the manuscript, as well as review and approval of the manuscript. Additional sponsors were not involved in manuscript preparation.

RESULTS

We enrolled 208 participants between November 2017 and December 2020. Of 208 cases, 51 (25%) were excluded from analysis because of insufficient tissue in the CB or absent CBs (Figure 1). All remaining 157 participants were female with a median age of 50 (IQR 41–61, range 25–87). All 157 cases were sufficient for ER/PR IHC, but 15/157 (10%) CBs were insufficient for HER2 testing, and 26/157 (17%) cases were insufficient for Ki-67 IHC. 95% (149/157) were categorised as malignant based on a diagnosis of carcinoma by histology, or if unavailable, then by cytology. If neither a histologic nor cytologic diagnosis was available, then a case was considered malignant if positive for HER2 by IHC and/or FISH. IHC performed on corresponding core biopsy or resections at MNH was available in 84/149 (56%) cases. Demographic and clinical characteristics of participants are summarised in Table 1.

Table 1.

Baseline Characteristics

Characteristic	Value	N (%)
Total		157 (100)
Age at diagnosis in years	Median (interquartile range)	50 (41,61)
	Mean (standard deviation)	52 (14)
Gender	Female	157 (100)
Estimated clinical size of breast mass	≤2 cm	6 (4)
	>2cm to ≤5cm	54 (34)
	> 5cm	87 (55)
	Unknown	10 (6)
Final diagnosis by FNAB	Adenocarcinoma	137 (87)
	Atypical	2 (1)
	Other benign diagnosis	3 (2)
	Unknown	15 (10)
Histologic tumor type based on subsequent histologic specimen	Invasive ductal carcinoma, not otherwise specified	95 (61)
	Mixed invasive ductal and lobular carcinoma	1 (<1)
	Invasive lobular carcinoma	1 (<1)
	Invasive papillary carcinoma	1 (<1)
	Medullary carcinoma	2 (1)
	Mucinous carcinoma	1 (<1)
	Non-diagnostic	3 (2)
	Other benign diagnoses	4 (4)
	Unknown	49 (31)
UCSF ER IHC status	Positive (≥10%)	81 (52)
	Low positive (1–9%)	1 (<1)
	Negative	75 (48)
UCSF PR IHC status	Positive (≥10%)	52 (33)
	Low positive (1–9%)	5 (3)
	Negative	100 (64)
UCSF HER2 IHC/FISH status	Positive, 3+	21 (13)
	Equivocal, 2+; HER2 FISH amplified	10 (6)
	Equivocal, 2+; HER2 FISH not amplified	0
	Negative, 1+	9 (6)
	Negative, 0	102 (65)
	Unknown	15 (10)
UCSF Ki-67 status	<20%	46 (29)
	≥20%	85 (54)
	Unknown	26 (17)
Malignant cases only ^
Total		149 (100)
UCSF ER IHC status	Positive (≥10%)	75 (50)
	Low positive (1–9%)	1 (<1)
	Negative	73 (49)
UCSF PR IHC status	Positive (≥10%)	46 (31)
	Low positive (1–9%)	5 (3)
	Negative	98 (66)
UCSF HER2 IHC/FISH status	Positive, 3+	21 (14)
	Equivocal, 2+; HER2 FISH amplified	10 (7)
	Equivocal, 2+; HER2 FISH not amplified	0 (0)
	Negative, 1+	9 (6)
	Negative, 0	95 (64)
	Unknown	14 (9)
UCSF Ki-67 status	<20%	42 (28)
	≥20%	83 (56)
	Unknown	24 (16)
MNH ER status	Positive	37 (25)
	Negative	47 (31)
	Unknown	65 (44)
MNH PR status	Positive	16 (11)
	Negative	68 (46)
	Unknown	65 (43)
MNH HER2 status	Positive	27 (18)
	Equivocal	4 (3)
	Negative	52 (35)
	Unknown	66 (44)
BC phenotype based on UCSF IHC and/or FISH	ER, PR+/−, HER2−	56 (38)
	ER+, PR+/−, HER2+	10 (7)
	ER-, PR+, HER2+	1 (<1)
	ER-, PR+, HER2-	1 (<1)
	ER−, PR−, HER2+	20 (13)
	ER+, PR+/−, HER2 unknown	10 (7)
	ER-, PR-, HER2 unknown	4 (3)
	TNBC	47 (32)
BC phenotype based on MNH IHC	ER+, PR+/−, HER2−	23 (15)
	ER+, PR+/−, HER2+	10 (7)
	ER-, PR+, HER2-	1 (<1)
	ER−, PR-, HER2+	17 (11)
	ER+, PR-, HER2 equivocal/unknown	4 (3)
	ER-, PR+, HER2 equivocal	1 (<1)
	TNBC	28 (19)
	Unknown	65 (44)

^{^}Includes only cases in which malignancy was confirmed by histologic evaluation, and if unavailable, then by fine needle aspiration. If neither a histologic nor cytologic diagnosis of carcinoma was available, then a case was included if it was HER2 positive by IHC and/or FISH.

Abbreviations: FNAB, fine needle aspiration biopsy; UCSF, University of California, San Francisco; ER, estrogen receptor; IHC, immunohistochemistry; PR, progesterone receptor; HER2, human epidermal growth factor receptor 2; FISH, fluorescence in situ hybridization; MNH, Muhimbili National Hospital; BC, breast cancer; TNBC, triple negative breast cancer

Because QL and M were introduced mid-study, not all samples were tested using all four preparation methods. 1x-FNAB was determined to be least practical and was discontinued once CAP validation guidelines were met (Figure 1).

Results of IHC performed at UCSF on CB specimens were compared with IHC performed at MNH on histologic specimens (appendix p 4). Of 149 cases, 84 (56%) cases had paired results for ER and PR IHC. Overall concordance was 85% (71/84) for ER (95% CI 77–92%), 74% (62/84) for PR (95% CI 64–83%), and 79% (55/70) for HER2 (95% CI 69–88%). Because FISH is not available at MNH, only definitive HER2 IHC results were compared, with 70 valid pairs. Based on the modest concordance and potential for variable processing of specimens at MNH, UCSF IHC and FISH were used as the reference standard for STRAT4 results.

All protocols tested had good to excellent concordance between STRAT4 ESR1 and ER IHC results (Table 2, Figures 2A-B, appendix p 5). 1x-FNAB smears performed the best out of the four protocols tested. Overall concordance of STRAT4 ESR1 results using 1x-FNAB was 95% [95% CI 90–100], sensitivity 94% [95% CI 85–100%], specificity 97% [95% CI 90–100%), and AUC of 0·96 (Figures 2A-B). Most of the discordant cases for 1x-CB (9/13, 69%) were false positive results, while most discordant cases for QL (12/13, 92%) and M (7/11, 64%) were false negative results. There was no evidence of imbalanced discordants for 1x-FNAB. For samples with a corresponding IHC result, the invalid rate was 2% (1/63) for 1x-FNAB, 17% (17/99) for QL, 18% (25/137) for M, 8% (12/155) for 1x-CB (appendix pp 6–10).

Table 2:

Agreement and utility of STRAT4 using 4 different protocols compared to UCSF IHC/FISH results

Preparation	Pairs N	Concordance % [95% CI] *	Kappa K [95% CI] *	McNemar p-value	Sensitivity % [95% CI] a	Specificity % [95% CI] *	PPV % [95% CI] *	NPV % [95% CI] *
1x-FNAB
ESR1	62	95.2 [89.8,100]	90.3 [79.7, 100]	0.56	93.8 [85.4, 100]	96.7 [90.2, 100]	96.8 [90.6, 100]	93.6 [84.9, 100]
PGR	61	83.6 [74.3,92.9]	63.5 [43.8, 83.2]	0.01	62.5 [43.1, 81.9]	97.3 [92.1, 100]	93.8 [81.9, 100]	80.0 [68.3, 91.7]
ERBB2	58	65.5 [53.3,77.8]	35.3 [16.6, 54.1]	<0.01	93.3 [80.7, 100]	55.8 [41.0, 70.7]	42.4 [25.6, 59.3]	96.0 [88.3, 100]
MKi67	53	64.2 [51.2,77.1]	26.7 [1.0, 52.8]	0.82	70.0 [53.6, 86.4]	56.5 [36.3, 76.8]	67.7 [51.3, 84.2]	59.1 [38.6, 79.6]
QL
ESR1	82	84.1 [76.2,92.1]	68.9 [53.9, 83.8]	<0.01	73.9 [61.2, 86.6]	97.2 [91.9, 100]	97.1 [91.6, 100]	74.5 [62.0, 86.9]
PGR	53	75.5 [63.9,87.1]	47.1 [24.9, 69.4]	<0.01	47.8 [27.4, 68.2]	96.7 [90.2, 100]	91.7 [76.0, 100]	70.7 [56.8, 84.7]
ERBB2	74	91.9 [85.7,98.1]	77.1 [59.7, 94.5]	0.41	87.5 [71.3, 100]	93.1 [86.6, 99.6]	77.8 [58.6, 97.0]	96.4 [91.6, 100]
MKi67	71	52.1 [40.5,63.7]	8.7 [−12.0, 29.4]	0.01	44.4 [29.9, 59.0]	65.4 [47.1, 83.7]	69.0 [52.1, 85.8]	40.5 [25.6, 55.3]
M
ESR1	112	90.2 [84.7,95.7]	80.4 [69.4, 91.4]	0.37	87.7 [79.2, 96.2]	92.7 [85.9, 99.6]	92.6 [85.6, 99.6]	87.9 [79.6, 96.3]
PGR	94	77.7 [69.2,86.1]	47.6 [29.6, 65.7]	<0.01	48.6 [32.0, 65.1]	94.9 [89.3, 100]	85.0 [69.4, 100]	75.7 [65.9, 85.5]
ERBB2	104	93.3 [88.5,98.1]	82.8 [70.5, 95.0]	0.06	96.0 [88.3, 100]	92.4 [86.6, 98.3]	80.0 [65.7, 94.3]	98.7 [96.0, 100]
MKi67	98	69.4 [60.3,78.5]	40.1 [23.6, 56.5]	<0.01	62.1 [50.4, 73.8]	84.4 [71.8, 97.0]	89.1 [80.1, 98.1]	51.9 [38.3, 65.5]
1x-CB
ESR1	143	90.9 [86.2,95.6]	81.7 [72.3, 91.2]	0.17	94.7 [89.6, 99.8]	86.8 [78.7, 94.8]	88.8 [81.8, 95.7]	93.7 [87.6, 99.7]
PGR	80	93.8 [88.5,99.1]	84.6 [71.6, 97.6]	0.18	83.3 [68.4, 98.2]	98.2 [94.8, 100]	95.2 [86.1, 100]	93.2 [86.8, 99.6]
ERBB2	131	89.3 [84.0,94.6]	72.5 [59.3, 85.7]	<0.01	96.4 [89.6, 100]	87.4 [81.0, 93.8]	67.5 [53.0, 82.0]	98.9 [96.8, 100]
MKi67	122	70.5 [62.4,78.6]	34.7 [17.3, 52.0]	0.74	78.5 [69.4, 87.5]	55.8 [41.0, 70.7]	76.5 [67.3, 85.8]	58.5 [43.5, 73.6]

^*Asymptotic normal standard error CI.

Abbreviations: 1x-FNAB, 1x protocol using fine needle aspiration biopsy smears; M, Maui protocol; QL, Quick lysis protocol; 1x-CB, 1x protocol using cell block material; CI, confidence interval; PPV, positive predictive value; NPV, negative predictive value

Figure 2.

Comparison of estrogen receptor and progesterone status as determined by STRAT4 and immunohistochemistry (IHC) for 1x protocol (1x-FNAB), quick lysis (QL), and Maui (M) on FNAB smears, and 1x on cell block sections (1x-CB). A) Box and whisker plots of STRAT4 ESR1 dCt values by ER IHC result categorised as negative (0%), low positive (1–9%), or positive (≥ 10%). B) Receiver operating characteristic curves for valid STRAT4 ESR1 results. C) Box and whisker plots of STRAT4 PGR dCt values by PR IHC result categorised as negative (0%), “low positive” (1–9%), or positive (≥ 10%). D) Receiver operating characteristic curves for STRAT4 PGR results.

For PR, 1x-CB had the highest overall agreement, while 1x-FNAB performed the best out of the smear preparations (Table 2, Figures 2C-D, appendix p 5). 1x-CB had overall concordance of 94% [95% CI 89–99%], sensitivity 83% [95% CI 68–98%)] specificity 98% [95% CI 95–100%], and AUC of 0·98 between STRAT4 PGR1 and PR IHC results. Excluding indeterminate cases, most of the discordant cases for all preparations were false negative results by STRAT4 (1x-FNAB, 90%, 9/10; 1x-CB, 80%, 4/5; M, 86%, 18/21; QL, 92%, 12/13) (appendix pp 11–15). Proportionally, 1x-CB had the most indeterminate results (41%, 63/155), while 1x-FNAB had the fewest (2%, 1/63). QL (29%, 29/99) had more indeterminate results than M (13%, 18/137). For samples with a corresponding IHC result, the invalid rate was 2% (1/63) for 1x-FNAB, 17% (17/99) for QL, 18% (25/137) for M, 8% (12/155) for 1x-CB (appendix pp 6–7).

For HER2, samples processed using M had the best performance when comparing STRAT4 ERBB2 with HER2 IHC/FISH results (Table 2, Figures 3A-B, appendix p 5). M had an overall concordance rate of 93% [95% CI 89–98%], sensitivity 96% [95% CI 88–100%], specificity 92% [95% CI 87–98%], and AUC of 0·95. For all preparations, most of the discordant cases were false positive results (1x-FNAB, 95%, 19/20; 1x-CB, 93%, 13/14; M, 86%, 6/7; QL, 67%, 4/6). There was only one sample with 2+ equivocal HER2 IHC and amplified ERBB2 FISH that rendered a false negative result using QL (appendix pp 16–18). All other cases with 2+ equivocal HER2 IHC but amplified HER2 FISH were correctly categorised as positive by STRAT4. For samples with a corresponding IHC result, the invalid rate was 2% (1/59) for 1x-FNAB, 14% (12/86) for QL, 16% (20/124) for M, 6% (9/140) for 1x-CB (appendix pp 6–7).

Figure 3.

Comparison of human epidermal growth factor receptor 2 status and Ki-67 proliferation index as determined by STRAT4 and immunohistochemistry (IHC) and/or fluorescence in situ hybridisation (FISH) for 1x protocol (1x-FNAB), quick lysis (QL), and Maui (M) on FNAB smears, and 1x on cell block sections (1x-CB). A) Box and whisker plots of STRAT4 ERBB2 dCt values by HER2 IHC result categorised as negative (0–1+), indeterminate (2+), or positive (3+). B) Receiver operating characteristic curves for STRAT4 ERBB2 results. C) Box and whisker plots of STRAT4 MKI67 dCt values by Ki-67 IHC result categorised as low proliferation index (<20%) or high proliferation index (≥ 20%). D) Receiver operating characteristic curves for STRAT4 MKI67 results.

Using a cutoff of 20% for Ki-67 IHC results, agreement between STRAT4 MKI67 and Ki-67 IHC results was poor to fair for all preparation methods (Table 2, Figures 3C-D, appendix appendix p 5). For all preparations, most of discordant cases were false negative results (1x-FNAB, 47%, 9/19; 1x-CB, 47%, 17/36; M, 83%, 25/30; QL, 74%, 25/34) (Figure 3C, appendix pp 5, 19–23). For samples with a corresponding IHC result, the invalid rate was 2% (1/54) for 1x-FNAB, 11% (8/72) for QL, 15% (17/115) for M, 5% (7/129) for 1x-CB (appendix pp 6–7).

DISCUSSION

In this study we assessed results of STRAT4 using four processing methods for FNAB samples (1x-FNAB, 1x-CB, M, and QL) as compared to IHC and/or FISH performed on FFPE CBs.

All preparation methods for STRAT4 yielded high sensitivity and specificity for ER status, with 1x-FNAB demonstrating the best performance. For PR, the performance for 1x-FNAB, QL, and M were modest. Although, the overall 1x-CB performance for PR was strong and consistent with prior studies, the indeterminate rate was 41%, limiting clinical utility of the assay. The indeterminate rate was higher than previously reported rate of 2·6% from a study using FFPE samples prepared in Rwanda.¹⁷ It is possible the CB preparation method, which is also an off-label sample type for the Xpert Breast Cancer STRAT4, affected the assay.

Performance of STRAT4 in determining HER2 status was excellent for M and 1x-CB, but the invalid rate for M is too high for clinical use. In addition, most of the discordant cases were false positives, which could plausibly result in inappropriate treatment with trastuzumab. It is unclear why FNAB smears were prone to false positive HER2 results, but this finding was previously noted in another study.¹⁸ One reason may be that the ERBB2 cutoff was developed using FFPE samples and not optimised for FNAB. Because there were only a few HER2 2+ cases, we are unable to draw conclusions about whether the STRAT4 can distinguish between cases that are amplified versus not amplified by ERBB2 FISH. However, all IHC 2+ cases tested were confirmed to be HER2-positive, based on having gene amplification by FISH, and these results were in agreement with STRAT4 “positive” results. Agreement between the STRAT4 and IHC results for Ki-67 was slight to fair, aligning with prior studies.^9,10 An optimal clinical cutoff for Ki-67 IHC has not yet been agreed on, and Ki-67 concordance can be modest between CBs and surgical resections.^19,20 Nevertheless, Ki-67 proliferation index is an integral component in determining breast cancer treatment, and additional optimisation to improve the performance of STRAT4 on all sample types is needed prior to clinical implementation. In addition, the distribution of molecular subtypes in this study is similar to the published literature from Tanzania.^12,21,22

For other studies that have evaluated STRAT4 using samples from SSA, overall concordance for ER was 90–93%, PR was 87–92%, HER2 was 93–98%, and Ki-67 was 82–84% using FFPE.^17,23,24 Of note, one of these studies used tissue processed in Rwanda, but specimens underwent biomarker testing in the United States. Our study was strengthened by performance of all processing and analyses in Tanzania, indicating that this technology can be implemented in a resource-constrained setting when appropriate technical training is availed.

To date, there has been only one previous study evaluating STRAT4 using FNAB specimens with a small sample size (N=20). The study from South Africa evaluated M and QL and reported overall concordance rates of 84% and 81% for ER, 94% and 81% for PR, 79% and 81% for HER2, and 67% and 63% for Ki-67, respectively.²⁴ Overall, our data appear comparable.

Limitations of this study should be acknowledged. Convenience sampling may introduce bias. Although we attempted to retrieve the pathology reports or clinical slides to document diagnoses, there were cases when neither were available; therefore, a subset of cases did not have documented pathologic diagnoses. The high rate of unknown diagnoses may also reflect loss to follow-up, since subsequent surgical resection or core needle biopsies were not available for evaluation. In addition, multiple new techniques were introduced to the Central Pathology Laboratory at MNH to support implementation of this study and to advance pathology capacity in Tanzania, including ROSE and formalin-based CBs.^6,8 Proficiency of the study technician improved with time, requiring an activation period with investment in training. FNAB is also highly operator dependent, but both pathologists had participated in dedicated training and were proficient at the outset.^25,26 Not all samples could be tested using all four preparation methods, because QL and M were introduced part-way through the study. 1x-FNAB was also felt to be the least practical and was discontinued once CAP validation guidelines were met. While FNAB cannot differentiate between in situ and invasive carcinoma, most patients in Tanzania present with advanced disease.^6,21

For 1x-CB, 42% of invalid cases (5/12) had scant cellularity which may contribute to test failure. For testing on FNAB smears, a majority had moderate to high cellularity, so test failure was unlikely related to cellularity. However, 58% (25/43) of the invalid smear results were performed >20 days after FNAB. Regarding discrepant cases, cellularity appeared to be only a minor issue since most cell blocks and smears had at least moderate cellularity. Smear age and quality of mRNA may have been a contributing factor and could yield both false positive and false negative results. However, many smears tested >20 days after collection still produced valid and concordant results.

This study was conducted by a highly trained team at a national referral hospital in Tanzania, which may not be representative of resources available in all healthcare settings. Future implementation and broader dissemination of STRAT4 using FNAB samples will require technical training. Both FNAB and ROSE require intensive training, which may constrain future scale-up.

This study demonstrates that pairing FNAB with STRAT4 is feasible in an LMIC and that concordance between STRAT4 and IHC for ER is robust. While findings demonstrate need for refinement of the assay for both ER and HER2, deployment of STRAT4 using FNAB for ER testing alone would have a profound impact on breast cancer care in LMICs by reducing diagnostic barriers linking patients to life-saving and life-prolonging endocrine therapies.^27,28 One study modeling implementation of STRAT4 in SSA projected that STRAT4 could provide moderate cost savings and reduce resource demands.²⁹ Further optimisation and testing of the STRAT4 and FNAB will provide critical information on performance in resource-constrained settings and will inform best practices for scaled implementation.

In conclusion, limited access to breast cancer biomarker testing is a major barrier to breast cancer care in resource-constrained settings. Histopathology and IHC are technically complex processes that require costly equipment and reagents and highly trained staff. By contrast, endocrine therapies are readily available at a low cost in LMICs. Therefore, development of leapfrog solutions for breast cancer biomarkers is a high priority globally.³⁰

RESEARCH IN CONTEXT

Evidence before this study

Poor outcomes for breast cancer in low and middle income countries (LMICs) is in part attributable to a lack of access to high quality pathology services. Accurate and readily accessible breast cancer biomarker testing for estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2) is critical for classification, guiding optimal breast cancer treatment, and determining prognosis. However, a recent survey of pathology laboratories in sub-Saharan Africa (SSA) found that only half provide onsite immunohistochemistry (IHC) services. The Xpert^® Breast Cancer STRAT4 assay (STRAT4) is a validated CE-marked in vitro diagnostic automated cartridge-based multiplex polymerase chain reaction assay that runs on the GeneXpert^® Instrument System and delivers semi-quantitative mRNA levels for ESR1, PGR, ERBB2 and MKI67 in formalin fixed paraffin embedded (FFPE) breast cancer tissue specimens. STRAT4 has been proposed as a possible solution to limitations in IHC in LMICs. Fine needle aspiration biopsy is the primary method of tissue sampling in many low resource settings. Retrospective studies have shown that FNAB has high sensitivity and specificity in diagnosing malignant masses when performed by trained medical personnel.

The performance of STRAT4 and IHC concordance has been shown to be robust and reproducible when using formalin-fixed paraffin embedded tissue in large studies performed in the United States and Europe (98–99% for ER, 90% for PR, 93–98% for HER2, and 79–85% for Ki-67. For studies evaluating STRAT4 using samples from SSA, the overall concordance for ER was 90–93%, PR was 87–92%, HER2 was 93–98%, and Ki-67 was 82–84% using formalin-fixed paraffin embedded tissue. To date, only one small pilot study from South Africa has published data assessing the feasibility of using FNAB samples for STRAT4. We searched PubMed using the terms (“STRAT4” OR “Xpert Breast Cancer STRAT4” OR “GeneXpert”) AND (“Breast cancer”), with no date or language restrictions.

Added value of this study

Our work represents one of the largest studies evaluating STRAT4 using FNAB specimens. We specifically assessed the use of FNAB samples because it can be deployed in settings with limited pathology infrastructure. We compared the performance of STRAT4 on FNAB specimens in Tanzania using 4 sample processing methods to IHC and/or fluorescence in situ hybridisation performed on FFPE cell blocks at an academic center in the United States. All preparation methods demonstrated high concordance between STRAT4 ESR1 and ER IHC using FNAB material. The results for the other biomarkers had more modest performance. Therefore, STRAT4 testing using FNAB samples in a resource constrained setting is feasible but further optimisation of the assay for PGR, ERBB2 and MKI67 is needed prior to clinical implementation of STRAT4.

Implications of all the available evidence

The high concordance between STRAT4 using FNAB samples for ER testing alone is promising. Increasing access to ER testing alone has the potential to transform breast cancer care in LMICs by significantly increasing the number of women eligible for lifesaving and readily available endocrine therapies, particularly in settings previously with no-to-limited access to pathology services. However, the findings also demonstrate opportunities for further develop and optimisation of STRAT4 for FNAB samples. Additional testing using FNAB samples is needed to inform development and best practices for implementing this technology.