Using a machine learning algorithm to detect depressed ejection fraction from a single-lead ECG

Background

Multiple studies demonstrate the benefit of intervention for left ventricular ejection fraction (LVEF) below 40%, so the development of a low ejection fraction algorithm to detect LVEF below 40% can aid in early screening of initial asymptomatic Heart Failure with reduced Ejection Fraction (HFrEF).

Objective

To demonstrate the performance of a low ejection fraction algorithm using single-lead ECG data to detect LVEF below 40%.

Methods

We collected 1325 single-lead ECG recordings (15s duration) at various chest positions using an electronic stethoscope from 197 patients. We analyzed these ECG recordings using a deep neural network model trained on individual leads extracted from a 12-lead ECG to discriminate left ventricular ejection fractions (EFs) above or below different thresholds. We compared the model output to ejection fraction measured using echocardiograms.

Results

Across all recordings from all patients, we obtained an AUROC of 0.89, with a sensitivity of 88% and specificity of 74% using a model output threshold of 0.35 (Figure 1). The AUROC of recordings taken at different orientations and stances ranged from 0.85 to 0.92 (Table 1), with a sensitivity of at least 78% and specificity of at least 66% at any orientation.

Conclusion

Using a single lead ECG measured by an electronic stethoscope and a deep neural network model, we were able to detect depressed ejection fraction (≤40%) with a sensitivity of 88% and specificity of 74%. This work demonstrates the utility of a low-cost electronic stethoscope and machine learning for early screening and detection of depressed left ventricular ejection fraction.

Funding Acknowledgement

Type of funding sources: Private company. Main funding source(s): Eko Health

Combining serum microRNA and CA-125 as prognostic indicators of preoperative surgical outcome in women with high-grade serous ovarian cancer

OBJECTIVES

The most widely used approach for the clinical management of women with high-grade serous ovarian cancer (HGSOC) is surgery, followed by platinum and taxane based chemotherapy. The degree of macroscopic disease remaining at the conclusion of surgery is a key prognostic factor determining progression free and overall survival. We sought to develop a non-invasive test to assist surgeons to determine the likelihood of achieving complete surgical resection. This knowledge could be used to plan surgical approaches for optimal clinical management.

METHODS

We profiled 170 serum microRNAs (miRNAs) using the Serum/Plasma Focus miRNA PCR panel containing locked nucleic acid (LNA) primers (Exiqon) in women with HGSOC (N=56) and age-matched healthy volunteers (N=30). Additionally, we measured serum CA-125 levels in the same samples. The HGSOC cohort was further classified based on the degree of macroscopic disease at the conclusion of surgery. Stepwise logistic regression was used to identify predictive markers.

RESULTS

We identified a combination of miR-375 and CA-125 as the strongest discriminator of healthy versus HGSOC serum, with an area under the curve (AUC) of 0.956. The inclusion of miR-210 increased the AUC to 0.984; however, miR-210 was affected by hemolysis. The combination of miR-34a-5p and CA-125 was the strongest predictor of completeness of surgical resection with an AUC of 0.818.

CONCLUSION

A molecular test incorporating circulating miRNA to predict completeness of surgical resection for women with HGSOC has the potential to contribute to planning for optimal patient management, ultimately improving patient outcome.

Lessons learnt from implementation of a Lynch syndrome screening program for patients with gynaecological malignancy

Despite a trend towards universal testing, best practice to screen patients presenting with gynaecological malignancy for Lynch syndrome (LS) is uncertain. We report our institutional experience of a co-ordinated gynaecological LS screening program. All patients with endometrial carcinoma or carcinosarcoma, or gynaecological endometrioid or clear cell carcinomas undergo reflex four panel immunohistochemistry (IHC) for MLH1, PMS2, MSH2 and MSH6 followed by cascade somatic hypermethylation analysis of the MLH1 promoter locus for dual MLH1/PMS2 negative tumours. On the basis of these results, genetic counselling and targeted germline mutation testing is then offered to patients considered at high risk of LS. From 1 August 2013 to 31 December 2015, 124 patients were screened (mean age 64.6 years). Thirty-six (29.0%) demonstrated abnormal MMR IHC: 26 (72.2%) showed dual loss of MLH1/PMS2, five (13.9%) dual loss of MSH2/MSH6, three (8.3%) isolated loss of MSH6, and two (5.6%) isolated loss of PMS2. Twenty-five of 26 (96.1%) patients with dual MLH1/PMS2 loss demonstrated MLH1 promoter methylation. Therefore, 11 (8.9%) patients screened were classified as high risk for LS, of whom nine (81.8%) accepted germline mutation testing. Three (2.4% of total screened) were confirmed to have LS, two with germline PMS2 and one with germline MSH2 mutation. Massive parallel sequencing of tumour tissue demonstrated somatic mutations which were concordant with the IHC results in the remainder. Interestingly, the one MLH1/PMS2 IHC negative but not hypermethylated tumour harboured only somatic MLH1 mutations, indicating that universal cascade methylation testing in MLH1/PMS2 IHC negative tumours is very low yield and could be reconsidered in a resource-poor setting. In conclusion, universal screening for LS in patients presenting with gynaecological malignancy using the algorithm described above identified LS in three of 124 (2.4%) of our population. Only three of nine (33.3%) patients considered at high risk for LS by combined IHC and hypermethylation analysis were proven to have LS. Only one of the LS patients was less than 50 years of age and none of these patients would have been identified had more restrictive Amsterdam or Bethesda criteria been applied.

Assessing mutant p53 in primary high-grade serous ovarian cancer using immunohistochemistry and massively parallel sequencing

The tumour suppressor p53 is mutated in cancer, including over 96% of high-grade serous ovarian cancer (HGSOC). Mutations cause loss of wild-type p53 function due to either gain of abnormal function of mutant p53 (mutp53), or absent to low mutp53. Massively parallel sequencing (MPS) enables increased accuracy of detection of somatic variants in heterogeneous tumours. We used MPS and immunohistochemistry (IHC) to characterise HGSOCs for TP53 mutation and p53 expression. TP53 mutation was identified in 94% (68/72) of HGSOCs, 62% of which were missense. Missense mutations demonstrated high p53 by IHC, as did 35% (9/26) of non-missense mutations. Low p53 was seen by IHC in 62% of HGSOC associated with non-missense mutations. Most wild-type TP53 tumours (75%, 6/8) displayed intermediate p53 levels. The overall sensitivity of detecting a TP53 mutation based on classification as ‘Low’, ‘Intermediate’ or ‘High’ for p53 IHC was 99%, with a specificity of 75%. We suggest p53 IHC can be used as a surrogate marker of TP53 mutation in HGSOC; however, this will result in misclassification of a proportion of TP53 wild-type and mutant tumours. Therapeutic targeting of mutp53 will require knowledge of both TP53 mutations and mutp53 expression.

Abstract B05: Assessment of TP53 mutation status in primary high-grade serous ovarian cancer and cell line models: Comparison between immunohistochemistry and next-generation sequencing

Mutations in TP53 have been shown to occur in around 96% of high-grade serous ovarian cancer (HGSOC). Different TP53 mutations can lead to accumulation of mutant p53 in the cell that support gain of function effects, or loss or abrogation of both wild-type and mutant p53. Success of new therapeutic strategies for HGSOC that target mutant p53 will likely depend on detailed knowledge of the type of TP53 mutation present and its influence on mutant p53 levels in the cell. The aim of this study was to identify the spectrum of TP53 mutations in a cohort of primary HGSOCs and compare this with p53 immunostaining patterns. In addition, we sought to identify TP53 mutations in the previously uncharacterised HGSOC cell line models OV167 and OV202, as well as the clear cell adenocarcinoma line OV207.

DNA from fresh frozen tumors and cell line samples were processed for next generation sequencing using The Fluidigm Access TP53 Array System run on the Fluidigm Biomark HD™ Real-Time PCR fluidics system (Fluidigm Corporation). Amplicon libraries were pooled and analyses performed in a single massively parallel sequencing run using the MiSeq Sequencing System (Illumina Inc.). Unaligned BAM files were analyzed within the Broad Institute's Genome Analysis Toolkit (GATK) best practice pipeline. The ANNOVAR software tool was used to generate an annotated variants list. Variants were further filtered based on their frequency in the 1000 Genome Database and predicted SIFT scores. Mutations were visualized using the Broad Institute's Integrative Genomics Viewer. All samples in which a mutation was identified underwent Sanger sequencing using primers recommended in the International Agency for Research on Cancer TP53 database. Immunohistochemistry for p53 was performed on formalin fixed paraffin embedded tissue sections using a commercially available mouse monoclonal anti-human antibody (Protein Clone DO-7, cat. #M7001, Dako) on an automated staining platform (Leica BOND-III™ autostainer). The percentage of cells showing positive nuclear staining for p53 was reported.

A mutation in TP53 was identified in 94% (68 of 72) of HGSOC. The majority of mutations identified (62%, 42 of 68) were missense, with all except one of these mutations located in the p53 DNA binding domain. Overall, 82% (56 of 68) of mutations occurred within the DNA binding domain. Frameshift, stop or splice mutations (38%, 26 of 68) were identified in exons or flanking intronic sequence between exons 4 – 10. The presence or location of a TP53 mutation did not influence overall survival. Sanger sequencing was unable to reliably detect variants at an allele frequency less than 25%. TP53 mutations were identified in all cell lines at a mutant allele frequency ≥99%.

Based on percent of nuclei positive for p53, tumors were immunohistochemically graded as High, Intermediate or Low for p53. All missense mutations scored High for expression of the p53 protein. Considerable variability was observed in p53 staining levels in samples containing stop, frameshift or splice mutations, with samples in all of High, Intermediate and Low categories. Furthermore, a single sample shown to be wild-type for TP53 was graded as High for p53, raising the possibility that p53 had been activated in this tumor via an alternative mechanism. The level of p53 did not affect overall survival.

Taken together, these results suggest that caution needs to be exercised when using p53 immunohistochemistry as a surrogate marker for either presence or type of TP53 mutation. A combination of next generation sequencing and p53 immunohistochemistry is required to robustly define both mutation type and effect on mutant p53 levels in the cell.

Citation Format: Alexander J. Cole, Trisha Dwight, Ying Zhu, Anthony J. Gill, Kristie-Ann Dickson, Adele Clarkson, Gregory B. Gard, Jayne Maidens, Susan Valmadre, Roderick J. Clifton-Bligh, Deborah J. Marsh. Assessment of TP53 mutation status in primary high-grade serous ovarian cancer and cell line models: Comparison between immunohistochemistry and next-generation sequencing. [abstract]. In: Proceedings of the AACR Special Conference on Advances in Ovarian Cancer Research: Exploiting Vulnerabilities; Oct 17-20, 2015; Orlando, FL. Philadelphia (PA): AACR; Clin Cancer Res 2016;22(2 Suppl):Abstract nr B05.

Elevated levels of circulating microRNA-200 family members correlate with serous epithelial ovarian cancer

Background

There is a critical need for improved diagnostic markers for high grade serous epithelial ovarian cancer (SEOC). MicroRNAs are stable in the circulation and may have utility as biomarkers of malignancy. We investigated whether levels of serum microRNA could discriminate women with high-grade SEOC from age matched healthy volunteers.

Methods

To identify microRNA of interest, microRNA expression profiling was performed on 4 SEOC cell lines and normal human ovarian surface epithelial cells. Total RNA was extracted from 500 μL aliquots of serum collected from patients with SEOC (n = 28) and age-matched healthy donors (n = 28). Serum microRNA levels were assessed by quantitative RT-PCR following preamplification.

Results

microRNA (miR)-182, miR-200a, miR-200b and miR-200c were highly overexpressed in the SEOC cell lines relative to normal human ovarian surface epithelial cells and were assessed in RNA extracted from serum as candidate biomarkers. miR-103, miR-92a and miR -638 had relatively invariant expression across all ovarian cell lines, and with small-nucleolar C/D box 48 (RNU48) were assessed in RNA extracted from serum as candidate endogenous normalizers. No correlation between serum levels and age were observed (age range 30-79 years) for any of these microRNA or RNU48. Individually, miR-200a, miR-200b and miR-200c normalized to serum volume and miR-103 were significantly higher in serum of the SEOC cohort (P < 0.05; 0.05; 0.0005 respectively) and in combination, miR-200b + miR-200c normalized to serum volume and miR-103 was the best predictive classifier of SEOC (ROC-AUC = 0.784). This predictive model (miR-200b + miR-200c) was further confirmed by leave one out cross validation (AUC = 0.784).

Conclusions

We identified serum microRNAs able to discriminate patients with high grade SEOC from age-matched healthy controls. The addition of these microRNAs to current testing regimes may improve diagnosis for women with SEOC.