Development and Evaluation of a Machine Learning Model for the Early Identification of Patients at Risk for Sepsis.

STUDY OBJECTIVE: The Third International Consensus Definitions (Sepsis-3) Task Force recommended the use of the quick Sequential [Sepsis-related] Organ Failure Assessment (qSOFA) score to screen patients for sepsis outside of the ICU. However, subsequent studies raise concerns about the sensitivity of qSOFA as a screening tool. We aim to use machine learning to develop a new sepsis screening tool, the Risk of Sepsis (RoS) score, and compare it with a slate of benchmark sepsis-screening tools, including the Systemic Inflammatory Response Syndrome, Sequential Organ Failure Assessment (SOFA), qSOFA, Modified Early Warning Score, and National Early Warning Score. METHODS: We used retrospective electronic health record data from adult patients who presented to 49 urban community hospital emergency departments during a 22-month period (N=2,759,529). We used the Rhee clinical surveillance criteria as our standard definition of sepsis and as the primary target for developing our model. The data were randomly split into training and test cohorts to derive and then evaluate the model. A feature selection process was carried out in 3 stages: first, we reviewed existing models for sepsis screening; second, we consulted with local subject matter experts; and third, we used a supervised machine learning called gradient boosting. Key metrics of performance included alert rate, area under the receiver operating characteristic curve, sensitivity, specificity, and precision. Performance was assessed at 1, 3, 6, 12, and 24 hours after an index time. RESULTS: The RoS score was the most discriminant screening tool at all time thresholds (area under the receiver operating characteristic curve 0.93 to 0.97). Compared with the next most discriminant benchmark (Sequential Organ Failure Assessment), RoS was significantly more sensitive (67.7% versus 49.2% at 1 hour and 84.6% versus 80.4% at 24 hours) and precise (27.6% versus 12.2% at 1 hour and 28.8% versus 11.4% at 24 hours). The sensitivity of qSOFA was relatively low (3.7% at 1 hour and 23.5% at 24 hours). CONCLUSION: In this retrospective study, RoS was more timely and discriminant than benchmark screening tools, including those recommend by the Sepsis-3 Task Force. Further study is needed to validate the RoS score at independent sites.

Development and Evaluation of an Automated Machine Learning Algorithm for In-Hospital Mortality Risk Adjustment Among Critical Care Patients.

OBJECTIVES: Risk adjustment algorithms for ICU mortality are necessary for measuring and improving ICU performance. Existing risk adjustment algorithms are not widely adopted. Key barriers to adoption include licensing and implementation costs as well as labor costs associated with human-intensive data collection. Widespread adoption of electronic health records makes automated risk adjustment feasible. Using modern machine learning methods and open source tools, we developed and evaluated a retrospective risk adjustment algorithm for in-hospital mortality among ICU patients. The Risk of Inpatient Death score can be fully automated and is reliant upon data elements that are generated in the course of usual hospital processes. SETTING: One hundred thirty-one ICUs in 53 hospitals operated by Tenet Healthcare. PATIENTS: A cohort of 237,173 ICU patients discharged between January 2014 and December 2016. DESIGN: The data were randomly split into training (36 hospitals), and validation (17 hospitals) data sets. Feature selection and model training were carried out using the training set while the discrimination, calibration, and accuracy of the model were assessed in the validation data set. MEASUREMENTS AND MAIN RESULTS: Model discrimination was evaluated based on the area under receiver operating characteristic curve; accuracy and calibration were assessed via adjusted Brier scores and visual analysis of calibration curves. Seventeen features, including a mix of clinical and administrative data elements, were retained in the final model. The Risk of Inpatient Death score demonstrated excellent discrimination (area under receiver operating characteristic curve = 0.94) and calibration (adjusted Brier score = 52.8%) in the validation dataset; these results compare favorably to the published performance statistics for the most commonly used mortality risk adjustment algorithms. CONCLUSIONS: Low adoption of ICU mortality risk adjustment algorithms impedes progress toward increasing the value of the healthcare delivered in ICUs. The Risk of Inpatient Death score has many attractive attributes that address the key barriers to adoption of ICU risk adjustment algorithms and performs comparably to existing human-intensive algorithms. Automated risk adjustment algorithms have the potential to obviate known barriers to adoption such as cost-prohibitive licensing fees and significant direct labor costs. Further evaluation is needed to ensure that the level of performance observed in this study could be achieved at independent sites.

Genome-wide association study in east Asians identifies novel susceptibility loci for breast cancer.

Genetic factors play an important role in the etiology of both sporadic and familial breast cancer. We aimed to discover novel genetic susceptibility loci for breast cancer. We conducted a four-stage genome-wide association study (GWAS) in 19,091 cases and 20,606 controls of East-Asian descent including Chinese, Korean, and Japanese women. After analyzing 690,947 SNPs in 2,918 cases and 2,324 controls, we evaluated 5,365 SNPs for replication in 3,972 cases and 3,852 controls. Ninety-four SNPs were further evaluated in 5,203 cases and 5,138 controls, and finally the top 22 SNPs were investigated in up to 17,423 additional subjects (7,489 cases and 9,934 controls). SNP rs9485372, near the TGF-ß activated kinase (TAB2) gene in chromosome 6q25.1, showed a consistent association with breast cancer risk across all four stages, with a P-value of 3.8×10(-12) in the combined analysis of all samples. Adjusted odds ratios (95% confidence intervals) were 0.89 (0.85-0.94) and 0.80 (0.75-0.86) for the A/G and A/A genotypes, respectively, compared with the genotype G/G. SNP rs9383951 (P = 1.9×10(-6) from the combined analysis of all samples), located in intron 5 of the ESR1 gene, and SNP rs7107217 (P = 4.6×10(-7)), located at 11q24.3, also showed a consistent association in each of the four stages. This study provides strong evidence for a novel breast cancer susceptibility locus represented by rs9485372, near the TAB2 gene (6q25.1), and identifies two possible susceptibility loci located in the ESR1 gene and 11q24.3, respectively.

Association of obesity-related genetic variants with endometrial cancer risk: a report from the Shanghai Endometrial Cancer Genetics Study.

Obesity is a well-established risk factor for endometrial cancer, the most common gynecologic malignancy. Recent genome-wide association studies (GWAS) have identified multiple genetic markers for obesity. The authors evaluated the association of obesity-related single nucleotide polymorphisms (SNPs) with endometrial cancer using GWAS data from their recently completed study, the Shanghai Endometrial Cancer Genetics Study, which comprised 832 endometrial cancer cases and 2,049 controls (1996-2005). Thirty-five SNPs previously associated with obesity or body mass index (BMI; weight (kg)/height (m)(2)) at a minimum significance level of ≤5 × 10(-7) in the US National Human Genome Research Institute's GWAS catalog (http://genome.gov/gwastudies) and representing 26 unique loci were evaluated by either direct genotyping or imputation. The authors found that for 22 of the 26 unique loci tested (84.6%), the BMI-associated risk variants were present at a higher frequency in cases than in population controls (P = 0.0003). Multiple regression analysis showed that 9 of 35 BMI-associated variants, representing 7 loci, were significantly associated (P ≤ 0.05) with the risk of endometrial cancer; for all but 1 SNP, the direction of association was consistent with that found for BMI. For consistent SNPs, the allelic odds ratios ranged from 1.15 to 1.29. These 7 loci are in the SEC16B/RASAL, TMEM18, MSRA, SOX6, MTCH2, FTO, and MC4R genes. The associations persisted after adjustment for BMI, suggesting that genetic markers of obesity provide value in addition to BMI in predicting endometrial cancer risk.

A Mendelian randomization analysis of circulating lipid traits and breast cancer risk.

BACKGROUND: Conventional epidemiologic studies have evaluated associations between circulating lipid levels and breast cancer risk, but results have been inconsistent. As Mendelian randomization analyses may provide evidence for causal inference, we sought to evaluate potentially unbiased associations between breast cancer risk and four genetically predicted lipid traits. METHODS: Previous genome-wide association studies (GWAS) have identified 164 discrete variants associated with high density lipoprotein-cholesterol (HDL-C), low density lipoprotein-cholesterol (LDL-C), triglycerides and total cholesterol. We used 162 of these unique variants to construct weighted genetic scores (wGSs) for a total of 101 424 breast cancer cases and 80 253 controls of European ancestry from the Breast Cancer Association Consortium (BCAC). Unconditional logistic regression was used to estimate odds ratios (OR) and 95% confidence intervals (CI) for associations between per standard deviation increase in genetically predicted lipid traits and breast cancer risk. Additional Mendelian randomization analysis approaches and sensitivity analyses were conducted to assess pleiotropy and instrument validity. RESULTS: Corresponding to approximately 15 mg/dL, one standard deviation increase in genetically predicted HDL-C was associated with a 12% increased breast cancer risk (OR: 1.12, 95% CI: 1.08-1.16). Findings were consistent after adjustment for breast cancer risk factors and were robust in several sensitivity analyses. Associations with genetically predicted triglycerides and total cholesterol were inconsistent, and no association for genetically predicted LDL-C was observed. CONCLUSIONS: This study provides strong evidence that circulating HDL-C may be associated with an increased risk of breast cancer, whereas LDL-C may not be related to breast cancer risk.

Molecular genetics of the platelet serotonin system in first-degree relatives of patients with autism.

Elevated platelet serotonin (5-hydroxytryptamine, 5-HT) is found in a subset of children with autism and in some of their first-degree relatives. Indices of the platelet serotonin system, including whole blood 5-HT, 5-HT binding affinity for the serotonin transporter (K(m)), 5-HT uptake (V(max)), and lysergic acid diethylamide (LSD) receptor binding, were previously studied in 24 first-degree relatives of probands with autism, half of whom were selected for elevated whole blood 5-HT levels. All subjects were then genotyped for selected polymorphisms at the SLC6A4, HTR7, HTR2A, ITGB3, and TPH1 loci. Previous studies allowed an a priori prediction of SLC6A4 haplotypes that separated the subjects into three groups that showed significantly different 5-HT binding affinity (K(m), p=0.005) and 5-HT uptake rate (V(max), p=0.046). Genotypes at four individual polymorphisms in SLC6A4 were not associated with platelet 5-HT indices. Haplotypes at SLC6A4 and individual genotypes of polymorphisms at SLC6A4, HTR7, HTR2A, ITGB3, and TPH1 showed no significant association with whole blood 5-HT. Haplotype analysis of two polymorphisms in TPH1 revealed a nominally significant association with whole blood 5-HT (p=0.046). These initial studies of indices of the 5-HT system with several single-nucleotide polymorphisms at loci in this system generate hypotheses for testing in other samples.

Variation in ITGB3 is associated with whole-blood serotonin level and autism susceptibility.

Autism is a pervasive developmental disorder affecting more males than females. Heritability estimates for autism can rise above 90%, and genes influencing the serotonin system are strong candidates for autism susceptibility genes, as drugs selectively acting on the serotonin system are some of the most effective treatments for maladaptive behaviors seen in autism. ITGB3 was recently identified as a male quantitative trait locus (QTL) for whole-blood serotonin levels in the Hutterites (P = 0.0003). Here, we demonstrate associations between variation in ITGB3 and serotonin levels in two outbred samples (P = 0.010 and 0.015). Lastly, we show that a coding variant of ITGB3 is associated with autism susceptibility in a large multiplex sample (P = 0.00082), and that this variation has different effects in males and females (P = 0.0018).

Beyond Epilepsy and Autism: Disruption of GABRB3 Causes Ocular Hypopigmentation.

Reduced ocular pigmentation is common in Angelman syndrome (AS) and Prader-Willi syndrome (PWS) and is long thought to be caused by OCA2 deletion. GABRB3 is located in the 15q11-13 region flanked by UBE3A, GABRA5, GABRG3, and OCA2. Mutations in GABRB3 have frequently been associated with epilepsy and autism, consistent with its role in neurodevelopment. We report here a robust phenotype in the mouse in which deletion of Gabrb3 alone causes nearly complete loss of retinal pigmentation due to atrophied melanosomes, as evidenced by electron microscopy. Using exome and RNA sequencing, we confirmed that only the Gabrb3 gene was disrupted while the Oca2 gene was intact. However, mRNA abundance of Oca2 and other genes adjacent to Gabrb3 is substantially reduced in Gabrb3-/- mice, suggesting complex transcriptional regulation in this region. These results suggest that impairment in GABRB3 downregulates OCA2 and indirectly causes ocular hypopigmentation and visual defects in AS and PWS.

Height and Breast Cancer Risk: Evidence From Prospective Studies and Mendelian Randomization.

BACKGROUND: Epidemiological studies have linked adult height with breast cancer risk in women. However, the magnitude of the association, particularly by subtypes of breast cancer, has not been established. Furthermore, the mechanisms of the association remain unclear. METHODS: We performed a meta-analysis to investigate associations between height and breast cancer risk using data from 159 prospective cohorts totaling 5216302 women, including 113178 events. In a consortium with individual-level data from 46325 case patients and 42482 control patients, we conducted a Mendelian randomization analysis using a genetic score that comprised 168 height-associated variants as an instrument. This association was further evaluated in a second consortium using summary statistics data from 16003 case patients and 41335 control patients. RESULTS: The pooled relative risk of breast cancer was 1.17 (95% confidence interval [CI] = 1.15 to 1.19) per 10cm increase in height in the meta-analysis of prospective studies. In Mendelian randomization analysis, the odds ratio of breast cancer per 10cm increase in genetically predicted height was 1.22 (95% CI = 1.13 to 1.32) in the first consortium and 1.21 (95% CI = 1.05 to 1.39) in the second consortium. The association was found in both premenopausal and postmenopausal women but restricted to hormone receptor-positive breast cancer. Analyses of height-associated variants identified eight new loci associated with breast cancer risk after adjusting for multiple comparisons, including three loci at 1q21.2, DNAJC27, and CCDC91 at genome-wide significance level P < 5×10(-8). CONCLUSIONS: Our study provides strong evidence that adult height is a risk factor for breast cancer in women and certain genetic factors and biological pathways affecting adult height have an important role in the etiology of breast cancer.

Genome-wide association analysis in East Asians identifies breast cancer susceptibility loci at 1q32.1, 5q14.3 and 15q26.1.

In a three-stage genome-wide association study among East Asian women including 22,780 cases and 24,181 controls, we identified 3 genetic loci newly associated with breast cancer risk, including rs4951011 at 1q32.1 (in intron 2 of the ZC3H11A gene; P=8.82×10(-9)), rs10474352 at 5q14.3 (near the ARRDC3 gene; P=1.67×10(-9)) and rs2290203 at 15q26.1 (in intron 14 of the PRC1 gene; P=4.25×10(-8)). We replicated these associations in 16,003 cases and 41,335 controls of European ancestry (P=0.030, 0.004 and 0.010, respectively). Data from the ENCODE Project suggest that variants rs4951011 and rs10474352 might be located in an enhancer region and transcription factor binding sites, respectively. This study provides additional insights into the genetics and biology of breast cancer.

Evaluating GWAS-identified SNPs for age at natural menopause among chinese women.

BACKGROUND: Age at natural menopause (ANM) is a complex trait with high heritability and is associated with several major hormonal-related diseases. Recently, several genome-wide association studies (GWAS), conducted exclusively among women of European ancestry, have discovered dozens of genetic loci influencing ANM. No study has been conducted to evaluate whether these findings can be generalized to Chinese women. METHODOLOGY/PRINCIPAL FINDINGS: We evaluated the index single nucleotide polymorphisms (SNPs) in 19 GWAS-identified genetic susceptibility loci for ANM among 3,533 Chinese women who had natural menopause. We also investigated 3 additional SNPs which were in LD with the index SNP in European-ancestry but not in Asian-ancestry populations. Two genetic risk scores (GRS) were calculated to summarize SNPs across multiple loci one for all SNPs tested (GRSall), and one for SNPs which showed association in our study (GRSsel). All 22 SNPs showed the same association direction as previously reported. Eight SNPs were nominally statistically significant with P≤0.05: rs4246511 (RHBDL2), rs12461110 (NLRP11), rs2307449 (POLG), rs12611091 (BRSK1), rs1172822 (BRSK1), rs365132 (UIMC1), rs2720044 (ASH2L), and rs7246479 (TMEM150B). Especially, SNPs rs4246511, rs365132, rs1172822, and rs7246479 remained significant even after Bonferroni correction. Significant associations were observed for GRS. Women in the highest quartile began menopause 0.7 years (P = 3.24×10(-9)) and 0.9 years (P = 4.61×10(-11)) later than those in the lowest quartile for GRSsel and GRSall, respectively. CONCLUSIONS: Among the 22 investigated SNPs, eight showed associations with ANM (P<0.05) in our Chinese population. Results from this study extend some recent GWAS findings to the Asian-ancestry population and may guide future efforts to identify genetic determination of menopause.

A common deletion in the APOBEC3 genes and breast cancer risk.

BACKGROUND: Genome-wide association studies (GWASs) have identified multiple genetic susceptibility loci for breast cancer. However, these loci explain only a small fraction of the heritability. Very few studies have evaluated copy number variation (CNV), another important source of human genetic variation, in relation to breast cancer risk. METHODS: We conducted a CNV GWAS in 2623 breast cancer patients and 1946 control subjects using data from Affymetrix SNP Array 6.0 (stage 1). We then replicated the most promising CNV using real-time quantitative polymerase chain reaction (qPCR) in an independent set of 4254 case patients and 4387 control subjects (stage 2). All subjects were recruited from population-based studies conducted among Chinese women in Shanghai. RESULTS: Of the 268 common CNVs (minor allele frequency ≥ 5%) investigated in stage 1, the strongest association was found for a common deletion in the APOBEC3 genes (P = 1.1×10(-4)) and was replicated in stage 2 (odds ratio =1.35, 95% confidence interval [CI] = 1.27 to 1.44; P = 9.6×10(-22)). Analyses of all samples from both stages using qPCR data produced odds ratios of 1.31 (95% CI = 1.21 to 1.42) for a one-copy deletion and 1.76 (95% CI = 1.57 to 1.97) for a two-copy deletion (P = 2.0×10(-24)). CONCLUSIONS: We provide convincing evidence for a novel breast cancer locus at the APOBEC3 genes. This CNV is one of the strongest common genetic risk variants identified so far for breast cancer.

No association between ovarian cancer susceptibility variants and breast cancer risk among Chinese women.

BACKGROUND: As breast and ovarian cancers may have similar etiologies, this study aimed to evaluate the hypothesis that breast cancer shares common genetic susceptibility variants with ovarian cancer. METHODS: Ten genetic variants in nine loci were previously identified to be associated with ovarian cancer risk among Caucasian women; an additional 353 variants in high-linkage disequilibrium (r(2) ≥ 0.6) among Han Chinese were identified. Data were available from the Affymetrix Genome-Wide Array (6.0) or MACH imputation for 25 and 78 common genetic variants [minor allele frequency (MAF) ≥0.05], respectively. Associations with breast cancer risk were evaluated by additive logistic regression models among 2,918 breast cancer cases and 2,324 controls. RESULTS: No associations with breast cancer risk were evident for 103 ovarian cancer susceptibility variants in five loci. Four loci were not evaluated, as they included only rare variants (MAF < 0.05). CONCLUSIONS: Ovarian cancer susceptibility variants identified in Caucasian women were not associated with breast cancer risk among 5,242 Chinese women. IMPACT: These findings suggest that breast and ovarian cancer may not share common susceptibility variants among Chinese women.

Genome-wide association analyses in East Asians identify new susceptibility loci for colorectal cancer.

To identify new genetic factors for colorectal cancer (CRC), we conducted a genome-wide association study in east Asians. By analyzing genome-wide data in 2,098 cases and 5,749 controls, we selected 64 promising SNPs for replication in an independent set of samples, including up to 5,358 cases and 5,922 controls. We identified four SNPs with association P values of 8.58 × 10(-7) to 3.77 × 10(-10) in the combined analysis of all east Asian samples. Three of the four were replicated in a study conducted in 26,060 individuals of European descent, with combined P values of 1.22 × 10(-10) for rs647161 (5q31.1), 6.64 × 10(-9) for rs2423279 (20p12.3) and 3.06 × 10(-8) for rs10774214 (12p13.32 near the CCND2 gene), derived from meta-analysis of data from both east Asian and European-ancestry populations. This study identified three new CRC susceptibility loci and provides additional insight into the genetics and biology of CRC.

Polymorphisms in inflammation pathway genes and endometrial cancer risk.

BACKGROUND: Experimental and epidemiologic evidence have suggested that chronic inflammation may play a critical role in endometrial carcinogenesis. METHODS: To investigate this hypothesis, a two-stage study was carried out to evaluate single-nucleotide polymorphisms (SNP) in inflammatory pathway genes in association with endometrial cancer risk. In stage I, 64 candidate pathway genes were identified and 4,542 directly genotyped or imputed SNPs were analyzed among 832 endometrial cancer cases and 2,049 controls, using data from the Shanghai Endometrial Cancer Genetics Study. Linkage disequilibrium of stage I SNPs significantly associated with endometrial cancer (P < 0.05) indicated that the majority of associations could be linked to one of 24 distinct loci. One SNP from each of the 24 loci was then selected for follow-up genotyping. Of these, 21 SNPs were successfully designed and genotyped in stage II, which consisted of 10 additional studies including 6,604 endometrial cancer cases and 8,511 controls. RESULTS: Five of the 21 SNPs had significant allelic odds ratios (ORs) and 95% confidence intervals (CI) as follows: FABP1, 0.92 (0.85-0.99); CXCL3, 1.16 (1.05-1.29); IL6, 1.08 (1.00-1.17); MSR1, 0.90 (0.82-0.98); and MMP9, 0.91 (0.87-0.97). Two of these polymorphisms were independently significant in the replication sample (rs352038 in CXCL3 and rs3918249 in MMP9). The association for the MMP9 polymorphism remained significant after Bonferroni correction and showed a significant association with endometrial cancer in both Asian- and European-ancestry samples. CONCLUSIONS: These findings lend support to the hypothesis that genetic polymorphisms in genes involved in the inflammatory pathway may contribute to genetic susceptibility to endometrial cancer. Impact statement: This study adds to the growing evidence that inflammation plays an important role in endometrial carcinogenesis.

Meta-analysis identifies common variants associated with body mass index in east Asians.

Multiple genetic loci associated with obesity or body mass index (BMI) have been identified through genome-wide association studies conducted predominantly in populations of European ancestry. We performed a meta-analysis of associations between BMI and approximately 2.4 million SNPs in 27,715 east Asians, which was followed by in silico and de novo replication studies in 37,691 and 17,642 additional east Asians, respectively. We identified ten BMI-associated loci at genome-wide significance (P < 5.0 × 10(-8)), including seven previously identified loci (FTO, SEC16B, MC4R, GIPR-QPCTL, ADCY3-DNAJC27, BDNF and MAP2K5) and three novel loci in or near the CDKAL1, PCSK1 and GP2 genes. Three additional loci nearly reached the genome-wide significance threshold, including two previously identified loci in the GNPDA2 and TFAP2B genes and a newly identified signal near PAX6, all of which were associated with BMI with P < 5.0 × 10(-7). Findings from this study may shed light on new pathways involved in obesity and demonstrate the value of conducting genetic studies in non-European populations.

Allelic heterogeneity at the serotonin transporter locus (SLC6A4) confers susceptibility to autism and rigid-compulsive behaviors.

Autism is a spectrum of neurodevelopmental disorders with a primarily genetic etiology exhibiting deficits in (1) development of language and (2) social relationships and (3) patterns of repetitive, restricted behaviors or interests and resistance to change. Elevated platelet serotonin (5-HT) in 20%-25% of cases and efficacy of selective 5-HT reuptake inhibitors (SSRIs) in treating anxiety, depression, and repetitive behaviors points to the 5-HT transporter (5-HTT; SERT) as a strong candidate gene. Association studies involving the functional insertion/deletion polymorphism in the promoter (5-HTTLPR) and a polymorphism in intron 2 are inconclusive, possibly because of phenotypic heterogeneity. Nonetheless, mounting evidence for genetic linkage of autism to the chromosome 17q11.2 region that harbors the SERT locus (SLC6A4) supports a genetic effect at or near this gene. We confirm recent reports of sex-biased genetic effects in 17q by showing highly significant linkage driven by families with only affected males. Association with common alleles fails to explain observed linkage; therefore, we hypothesized that preferential transmission of multiple alleles does explain it. From 120 families, most contributing to linkage at 17q11.2, we found four coding substitutions at highly conserved positions and 15 other variants in 5' noncoding and other intronic regions transmitted in families exhibiting increased rigid-compulsive behaviors. In the aggregate, these variants show significant linkage to and association with autism. Our data provide strong support for a collection of multiple, often rare, alleles at SLC6A4 as imposing risk of autism.