In this study, we showed by means of a defined analytical validation and verification process developed according to the CLSI guidelines that the RT-qPCR-based EndoPredict multianalyte gene expression test is a robust test that can be performed reproducibly and accurately. The resulting performance characteristics therefore meet the requirements needed for a diagnostic test. Moreover, we verified that a comparable performance with respect to assay efficiency, precision, and accuracy can also be achieved in a routine molecular diagnostic laboratory. In addition, this study provides the specifications for analytical verification of EndoPredict in molecular pathological laboratories.
Successful clinical validation of the EndoPredict score in two large clinical trials was published previously
 resulting in a level of evidence of 1B according to the classification for prognostic biomarkers that has been proposed by Simon et al.
. The clinical validation studies which were performed within a prospective-retrospective design showed that the test predicted distant metastasis in patients with primary ER-positive, HER2-negative breast cancer and provided significant prognostic information beyond standard parameters to assess the risk of metastasis. Both, clinical and analytical validations of the EndoPredict multigene assay now fulfill the recently published recommendations for translating omics-based tests from research laboratory to clinical practice
The difficulty of a comprehensive analytical validation of a highly complex genomic multianalyte assay such as EndoPredict is that there are no evaluation guidelines to cover all relevant aspects. Therefore, we specified a validation program for both, the 12 single analyte assays and the combined multigene assay. Where applicable, established guidelines of the CLSI as well as the MIQE (Minimum Information for Publication of Quantitative Real-Time PCR Experiments) guidelines were followed
[13, 18]. Therefore, the analytical validation process conformed to the standard assay validation formats as close as reasonably possible.
Besides EndoPredict, other prognostic multigene expression tests for patients with breast or other cancers like MammaPrint
, Oncotype DX Breast Cancer or Colon Cancer Assays
[9, 10], or PAM50
[11, 12] are commercially available. Analytical performance characteristics of these tests, which must be performed in reference laboratories, were published only for some tests such as MammaPrint
 or the two Oncotype DX assays
[25, 26]. Here, we assessed the analytical performance data of EndoPredict and provide further evidence that EndoPredict is suitable for decentralized testing in specialized molecular pathological laboratories which is a unique feature not shown for the other RNA-based multigene expression tests. On one hand this conclusion is supported by the verification of the performance characteristics and the accuracy of the measurements in a routine laboratory. On the other hand, the data of the precision study in two different laboratories showed a total variation (standard deviation) of the EP score of 0.15 and 0.18, respectively, which is about 1% of the total EP score range from 0 to 15. This is similar to the variation of the Oncotype DX Breast Cancer Assay performed in one centralized company laboratory
. Interestingly, the main factor influencing total noise was not one of the individual variables tested, such as operator, PCR machine, day, day-time, or reagent lots but rather replicate-to-replicate noise which is assumed to be caused by PCR technique-inherent noise. In order to account for PCR-inherent noise the EndoPredict algorithm includes a “noise model” estimating the acceptable variance of replicate noise with respect to the Cq value. On this basis outlier elimination is accomplished
; one of its principal requirements is triplicate measurements.
Robust performance of the EndoPredict test in a decentralized setting was also shown in a successful prospective proficiency testing study including seven different molecular pathology laboratories
. In this study, 69 out of 70 EndoPredict measurements were within the pre-specified range, 100% of the samples were correctly classified as low or high risk of metastasis, and the total variation (standard deviation) of all measurements was 0.25 units, corresponding to 1.7% of the whole range of the EP score. The slightly larger total variation in comparison to the results from the analytical precision study where large RNA pools were used might be due to the fact that in the proficiency testing the participants received tissue sections and not RNA for testing, thus including variation induced by tumor heterogeneity.
A recent study assessed the variability of Ki-67 immunohistochemistry, which is a standard antibody-based diagnostic test in pathology used for treatment decision making in luminal breast cancer
. The authors found standard deviations of Ki-67 results obtained by 15 pathologists on centrally stained slides of three breast carcinomas ranging from 21.7% to 24.1%. Interestingly, even clear guidelines how to assess Ki-67 could not improve variability.
Although our results might suggest a higher reproducibility of the PCR-based test using standardized instruments and reagents, it is important to know that the pathological laboratories involved in this technical verification study as well in the proficiency testing of EndoPredict were highly experienced in molecular work. Therefore, the results might be different in laboratories with less molecular diagnostic experience and ongoing quality control by periodical round robin tests might be reasonable.
A critical issue for accurate results from any diagnostic assay is the use of optimal sample material. For this validation study optimal FFPE tumor material was selected by experienced pathologists on the basis of adjacent HE-stained tissue slides. Hence, in order to obtain high-quality EndoPredict results an expert pathological evaluation of the tissue specimens is mandatory. Moreover, a highly standardized method for RNA isolation as used in this validation study and in the EndoPredict proficiency testing is necessary. In this study, the validation of the RNA extraction method was not a primary aim. The silica-coated magnetic beads-based method used and recommended for EndoPredict was thoroughly validated in previous studies showing a reliable analysis of RNA transcript levels by RT-qPCR in FFPE tissue
[4–6]. Finally, the PCR platform used in this validation study was validated and CE-marked for diagnostic purposes by the manufacturer. It also supports robust performance of the EndoPredict assay. Therefore, change of isolation method, enzymes, and PCR platform might alter performance characteristics of this multianalyte assay.