Machine Learning for Prediction of Ventricular Arrhythmia Episodes from Intracardiac Electrograms of Automatic Implantable Cardioverter-Defibrillators.

BACKGROUND: Despite the implantable cardioverter defibrillator's (ICD) effectiveness in saving patients with life-threatening ventricular arrhythmias (VAs), the temporal occurrence of VA following ICD implantation is unpredictable. OBJECTIVE: Apply machine learning (ML) to intracardiac electrograms (IEGMs) recorded by ICDs as a unique biomarker for predicting impending VAs. METHODS: The study included 13,516 patients who received BIOTRONIK ICDs and enrolled in the CERTITUDE registry between 01/01/2010 to 12/31/2020. Database extraction included IEGMs from standard quarterly transmissions and VA event episodes. The processed IEGM data were pulled from device transmissions stored in a centralized Home Monitoring Service Center and reformatted into an analyzable format. Long- (baseline or first scheduled remote recording), mid-(scheduled remote recording every 90 days), or short-range predictions (IEGM within 5 seconds before the VA onset) were used to determine whether ML-processed IEGMs predicted impending VA events. Convolutional neural network classifiers using ResNet architecture were employed. RESULTS: Of 13,516 patients (male 72%, age 67.5 ± 11.9 years), 301,647 IEGM recordings were collected; 27,845 episodes of sustained VT/VF were observed in 4,467 patients (33.0%). Neural networks based on CNN using ResNet-like architectures on far-field IEGMs yielded an AUC of 0.83 with a 95% confidence interval of [0.79, 0.87] in the short-term, while the long- and mid-range analyses had minimal predictive value for VA events. CONCLUSION: In this study, applying ML to ICD-acquired IEGMs predicted impending VT/VF events seconds before they occurred, whereas mid- to long-term predictions were not successful. This could have important implications for future device therapies.

Spectrum of Response to Platinum and PARP Inhibitors in Germline BRCA-Associated Pancreatic Cancer in the Clinical and Preclinical Setting.

Germline BRCA-associated pancreatic ductal adenocarcinoma (glBRCA PDAC) tumors are susceptible to platinum and PARP inhibition. The clinical outcomes of 125 patients with glBRCA PDAC were stratified based on the spectrum of response to platinum/PARP inhibition: (i) refractory [overall survival (OS) <6 months], (ii) durable response followed by acquired resistance (OS <36 months), and (iii) long-term responders (OS >36 months). Patient-derived xenografts (PDX) were generated from 25 patients with glBRCA PDAC at different clinical time points. Response to platinum/PARP inhibition in vivo and ex vivo culture (EVOC) correlated with clinical response. We deciphered the mechanisms of resistance in glBRCA PDAC and identified homologous recombination (HR) proficiency and secondary mutations restoring partial functionality as the most dominant resistant mechanism. Yet, a subset of HR-deficient (HRD) patients demonstrated clinical resistance. Their tumors displayed basal-like molecular subtype and were more aneuploid. Tumor mutational burden was high in HRD PDAC and significantly higher in tumors with secondary mutations. Anti-PD-1 attenuated tumor growth in a novel humanized glBRCA PDAC PDX model. This work demonstrates the utility of preclinical models, including EVOC, to predict the response of glBRCA PDAC to treatment, which has the potential to inform time-sensitive medical decisions. SIGNIFICANCE: glBRCA PDAC has a favorable response to platinum/PARP inhibition. However, most patients develop resistance. Additional treatment options for this unique subpopulation are needed. We generated model systems in PDXs and an ex vivo system (EVOC) that faithfully recapitulate these specific clinical scenarios as a platform to investigate the mechanisms of resistance for further drug development. This article is highlighted in the In This Issue feature, p. 1749.

Prediction of Influenza Complications: Development and Validation of a Machine Learning Prediction Model to Improve and Expand the Identification of Vaccine-Hesitant Patients at Risk of Severe Influenza Complications.

Influenza vaccinations are recommended for high-risk individuals, but few population-based strategies exist to identify individual risks. Patient-level data from unvaccinated individuals, stratified into retrospective cases (n = 111,022) and controls (n = 2,207,714), informed a machine learning model designed to create an influenza risk score; the model was called the Geisinger Flu-Complications Flag (GFlu-CxFlag). The flag was created and validated on a cohort of 604,389 unique individuals. Risk scores were generated for influenza cases; the complication rate for individuals without influenza was estimated to adjust for unrelated complications. Shapley values were used to examine the model's correctness and demonstrate its dependence on different features. Bias was assessed for race and sex. Inverse propensity weighting was used in the derivation stage to correct for biases. The GFlu-CxFlag model was compared to the pre-existing Medial EarlySign Flu Algomarker and existing risk guidelines that describe high-risk patients who would benefit from influenza vaccination. The GFlu-CxFlag outperformed other traditional risk-based models; the area under curve (AUC) was 0.786 [0.783−0.789], compared with 0.694 [0.690−0.698] (p-value < 0.00001). The presence of acute and chronic respiratory diseases, age, and previous emergency department visits contributed most to the GFlu-CxFlag model's prediction. When higher numerical scores were assigned to more severe complications, the GFlu-CxFlag AUC increased to 0.828 [0.823−0.833], with excellent discrimination in the final model used to perform the risk stratification of the population. The GFlu-CxFlag can better identify high-risk individuals than existing models based on vaccination guidelines, thus creating a population-based risk stratification for individual risk assessment and deployment in vaccine hesitancy reduction programs in our health system.

Machine Learning for Early Lung Cancer Identification Using Routine Clinical and Laboratory Data.

Rationale: Most lung cancers are diagnosed at an advanced stage. Presymptomatic identification of high-risk individuals can prompt earlier intervention and improve long-term outcomes. Objectives: To develop a model to predict a future diagnosis of lung cancer on the basis of routine clinical and laboratory data by using machine learning. Methods: We assembled data from 6,505 case patients with non-small cell lung cancer (NSCLC) and 189,597 contemporaneous control subjects and compared the accuracy of a novel machine learning model with a modified version of the well-validated 2012 Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial risk model (mPLCOm2012), by using the area under the receiver operating characteristic curve (AUC), sensitivity, and diagnostic odds ratio (OR) as measures of model performance. Measurements and Main Results: Among ever-smokers in the test set, a machine learning model was more accurate than the mPLCOm2012 for identifying NSCLC 9-12 months before clinical diagnosis (P < 0.00001) and demonstrated an AUC of 0.86, a diagnostic OR of 12.3, and a sensitivity of 40.1% at a predefined specificity of 95%. In comparison, the mPLCOm2012 demonstrated an AUC of 0.79, an OR of 7.4, and a sensitivity of 27.9% at the same specificity. The machine learning model was more accurate than standard eligibility criteria for lung cancer screening and more accurate than the mPLCOm2012 when applied to a screening-eligible population. Influential model variables included known risk factors and novel predictors such as white blood cell and platelet counts. Conclusions: A machine learning model was more accurate for early diagnosis of NSCLC than either standard eligibility criteria for screening or the mPLCOm2012, demonstrating the potential to help prevent lung cancer deaths through early detection.

Leveraging Machine Learning Techniques to Forecast Patient Prognosis After Percutaneous Coronary Intervention.

OBJECTIVES: This study sought to determine whether machine learning can be used to better identify patients at risk for death or congestive heart failure (CHF) rehospitalization after percutaneous coronary intervention (PCI). BACKGROUND: Contemporary risk models for event prediction after PCI have limited predictive ability. Machine learning has the potential to identify complex nonlinear patterns within datasets, improving the predictive power of models. METHODS: We evaluated 11,709 distinct patients who underwent 14,349 PCIs between January 2004 and December 2013 in the Mayo Clinic PCI registry. Fifty-two demographic and clinical parameters known at the time of admission were used to predict in-hospital mortality and 358 additional variables available at discharge were examined to identify patients at risk for CHF readmission. For each event, we trained a random forest regression model (i.e., machine learning) to estimate the time-to-event. Eight-fold cross-validation was used to estimate model performance. We used the predicted time-to-event as a score, generated a receiver-operating characteristic curve, and calculated the area under the curve (AUC). Model performance was then compared with a logistic regression model using pairwise comparisons of AUCs and calculation of net reclassification indices. RESULTS: The predictive algorithm identified a high-risk cohort representing 2% of all patients who had an in-hospital mortality of 45.5% (95% confidence interval: 43.5% to 47.5%) compared with a risk of 2.1% for the general population (AUC: 0.925; 95% confidence interval: 0.92 to 0.93). Advancing age, CHF, and shock on presentation were the leading predictors for the outcome. A high-risk group representing 1% of all patients was identified with 30-day CHF rehospitalization of 8.1% (95% confidence interval: 6.3% to 10.2%). Random forest regression outperformed logistic regression for predicting 30-day CHF readmission (AUC: 0.90 vs. 0.85; p = 0.003; net reclassification improvement: 5.14%) and 180-day cardiovascular death (AUC: 0.88 vs. 0.81; p = 0.02; net reclassification improvement: 0.02%). CONCLUSIONS: Random forest regression models (machine learning) were more predictive and discriminative than standard regression methods at identifying patients at risk for 180-day cardiovascular mortality and 30-day CHF rehospitalization, but not in-hospital mortality. Machine learning was effective at identifying subgroups at high risk for post-procedure mortality and readmission.

Correction to: Early Colorectal Cancer Detected by Machine Learning Model Using Gender, Age, and Complete Blood Count Data.

The article Early Colorectal Cancer Detected by Machine Learning Model Using Gender, Age, and Complete Blood Count Data, written by Mark C. Hornbrook, Ran Goshen, Eran Choman, Maureen O'Keeffe-Rosetti, Yaron Kinar, Elizabeth G. Liles, and Kristal C. Rust, was originally published Online First without open access.

Early Colorectal Cancer Detected by Machine Learning Model Using Gender, Age, and Complete Blood Count Data.

BACKGROUND: Machine learning tools identify patients with blood counts indicating greater likelihood of colorectal cancer and warranting colonoscopy referral. AIMS: To validate a machine learning colorectal cancer detection model on a US community-based insured adult population. METHODS: Eligible colorectal cancer cases (439 females, 461 males) with complete blood counts before diagnosis were identified from Kaiser Permanente Northwest Region's Tumor Registry. Control patients (n = 9108) were randomly selected from KPNW's population who had no cancers, received at ≥1 blood count, had continuous enrollment from 180 days prior to the blood count through 24 months after the count, and were aged 40-89. For each control, one blood count was randomly selected as the pseudo-colorectal cancer diagnosis date for matching to cases, and assigned a "calendar year" based on the count date. For each calendar year, 18 controls were randomly selected to match the general enrollment's 10-year age groups and lengths of continuous enrollment. Prediction performance was evaluated by area under the curve, specificity, and odds ratios. RESULTS: Area under the receiver operating characteristics curve for detecting colorectal cancer was 0.80 ± 0.01. At 99% specificity, the odds ratio for association of a high-risk detection score with colorectal cancer was 34.7 (95% CI 28.9-40.4). The detection model had the highest accuracy in identifying right-sided colorectal cancers. CONCLUSIONS: ColonFlag® identifies individuals with tenfold higher risk of undiagnosed colorectal cancer at curable stages (0/I/II), flags colorectal tumors 180-360 days prior to usual clinical diagnosis, and is more accurate at identifying right-sided (compared to left-sided) colorectal cancers.

Predicting the presence of colon cancer in members of a health maintenance organisation by evaluating analytes from standard laboratory records.

BACKGROUND: A valid risk prediction model for colorectal cancer (CRC) could be used to identify individuals in the population who would most benefit from CRC screening. We evaluated the potential for information derived from a panel of blood tests to predict a diagnosis of CRC from 1 month to 3 years in the future. METHODS: We abstracted information on 1755 CRC cases and 54 730 matched cancer-free controls who had one or more blood tests recorded in the electronic records of Maccabi Health Services (MHS) during the period 30-180 days before diagnosis. A scoring model (CRC score) was constructed using the study subjects' blood test results. We calculated the odds ratio for being diagnosed with CRC after the date of blood draw, according to CRC score and time from blood draw. RESULTS: The odds ratio for having CRC detected within 6 months for those with a score of four or greater (vs three or less) was 7.3 (95% CI: 6.3-8.5) for men and was 7.8 (95% CI: 6.7-9.1) for women. CONCLUSIONS: Information taken from routine blood tests can be used to predict the risk of being diagnosed with CRC in the near future.

Performance analysis of a machine learning flagging system used to identify a group of individuals at a high risk for colorectal cancer.

Individuals with colorectal cancer (CRC) have a tendency to intestinal bleeding which may result in mild to severe iron deficiency anemia, but for many colon cancer patients hematological abnormalities are subtle. The fecal occult blood test (FOBT) is used as a pre-screening test whereby those with a positive FOBT are referred to colonscopy. We sought to determine if information contained in the complete blood count (CBC) report coud be processed automatically and used to predict the presence of occult colorectal cancer (CRC) in the setting of a large health services plan. Using the health records of the Maccabi Health Services (MHS) we reviewed CBC reports for 112,584 study subjects of whom 133 were diagnosed with CRC in 2008 and analysed these with the MeScore tool. The odds ratio for being diagnosed with CRC in 2008 was calculated with regards to the MeScore, using cutoff levels of 97% and 99% percentiles. For individuals in the highest one percentile, the odds ratio for CRC was 21.8 (95% CI 13.8 to 34.2). For the majority of the individuals with cancer, CRC was not suspected at the time of the blood draw. Frequent use of anticoagulants, the presence of other gastrointestinal pathologies and non-GI malignancies were assocaitged with false positive MeScores. The MeScore can help identify individuals in the population who would benefit most from CRC screening, including those with no clinical signs or symptoms of CRC.

Development and validation of a predictive model for detection of colorectal cancer in primary care by analysis of complete blood counts: a binational retrospective study.

OBJECTIVE: The use of risk prediction models grows as electronic medical records become widely available. Here, we develop and validate a model to identify individuals at increased risk for colorectal cancer (CRC) by analyzing blood counts, age, and sex, then determine the model's value when used to supplement conventional screening. MATERIALS AND METHODS: Primary care data were collected from a cohort of 606 403 Israelis (of whom 3135 were diagnosed with CRC) and a case control UK dataset of 5061 CRC cases and 25 613 controls. The model was developed on 80% of the Israeli dataset and validated using the remaining Israeli and UK datasets. Performance was evaluated according to the area under the curve, specificity, and odds ratio at several working points. RESULTS: Using blood counts obtained 3-6 months before diagnosis, the area under the curve for detecting CRC was 0.82 ± 0.01 for the Israeli validation set. The specificity was 88 ± 2% in the Israeli validation set and 94 ± 1% in the UK dataset. Detecting 50% of CRC cases, the odds ratio was 26 ± 5 and 40 ± 6, respectively, for a false-positive rate of 0.5%. Specificity for 50% detection was 87 ± 2% a year before diagnosis and 85 ± 2% for localized cancers. When used in addition to the fecal occult blood test, our model enabled more than a 2-fold increase in CRC detection. DISCUSSION: Comparable results in 2 unrelated populations suggest that the model should generally apply to the detection of CRC in other groups. The model's performance is superior to current iron deficiency anemia management guidelines, and may help physicians to identify individuals requiring additional clinical evaluation. CONCLUSIONS: Our model may help to detect CRC earlier in clinical practice.

Adenosine-to-inosine RNA editing shapes transcriptome diversity in primates.

Human and chimpanzee genomes are almost identical, yet humans express higher brain capabilities. Deciphering the basis for this superiority is a long sought-after challenge. Adenosine-to-inosine (A-to-I) RNA editing is a widespread modification of the transcriptome. The editing level in humans is significantly higher compared with nonprimates, due to exceptional editing within the primate-specific Alu sequences, but the global editing level of nonhuman primates has not been studied so far. Here we report the sequencing of transcribed Alu sequences in humans, chimpanzees, and rhesus monkeys. We found that, on average, the editing level in the transcripts analyzed is higher in human brain compared with nonhuman primates, even where the genomic Alu structure is unmodified. Correlated editing is observed for pairs and triplets of specific adenosines along the Alu sequences. Moreover, new editable species-specific Alu insertions, subsequent to the human-chimpanzee split, are significantly enriched in genes related to neuronal functions and neurological diseases. The enhanced editing level in the human brain and the association with neuronal functions both hint at the possible contribution of A-to-I editing to the development of higher brain function. We show here that combinatorial editing is the most significant contributor to the transcriptome repertoire and suggest that Alu editing adapted by natural selection may therefore serve as an alternate information mechanism based on the binary A/I code.

Systematic antigenic profiling of hematopoietic antigens on ovarian carcinoma cells identifies membrane proteins for targeted therapy development.

OBJECTIVE: We studied ovarian cancers for the expression of membrane markers of hematopoietic origin. STUDY DESIGN: We used flow cytometry to systematically characterize the expression of more than 30 hematologic antigens on ovarian carcinoma cell lines and to assess their stability under estrogen exposure. The expression of the antigens was validated by a bioinformatics survey and immunohistochemical staining of ovarian cancer specimens. RESULTS: Several antigens were expressed by the majority of the cells, such as CD15, CD71, and CD138, whereas others were found on small and distinct cellular subpopulations. The expression patterns of the different markers were unaffected by estrogen exposure, indicating their stability. CONCLUSION: The antigens described in our work may serve as potential targets for new and existing targeted drugs.

Evolutionarily conserved human targets of adenosine to inosine RNA editing.

A-to-I RNA editing by ADARs is a post-transcriptional mechanism for expanding the proteomic repertoire. Genetic recoding by editing was so far observed for only a few mammalian RNAs that are predominantly expressed in nervous tissues. However, as these editing targets fail to explain the broad and severe phenotypes of ADAR1 knockout mice, additional targets for editing by ADARs were always expected. Using comparative genomics and expressed sequence analysis, we identified and experimentally verified four additional candidate human substrates for ADAR-mediated editing: FLNA, BLCAP, CYFIP2 and IGFBP7. Additionally, editing of three of these substrates was verified in the mouse while two of them were validated in chicken. Interestingly, none of these substrates encodes a receptor protein but two of them are strongly expressed in the CNS and seem important for proper nervous system function. The editing pattern observed suggests that some of the affected proteins might have altered physiological properties leaving the possibility that they can be related to the phenotypes of ADAR1 knockout mice.

Is abundant A-to-I RNA editing primate-specific?

A-to-I RNA editing is common in all eukaryotes, and is associated with various neurological functions. Recently, A-to-I editing was found to occur frequently in the human transcriptome. In this article, we show that the frequency of A-to-I editing in humans is at least an order of magnitude higher than in the mouse, rat, chicken or fly genomes. The extraordinary frequency of RNA editing in human is explained by the dominance of the primate-specific Alu element in the human transcriptome, which increases the number of double-stranded RNA substrates.