Increasingly, cognitive scientists have demonstrated interest inapplying tools from deep learning. One use for deep learning isin language acquisition where it is useful to know if a linguisticphenomenon can be learned through domain-general means.To assess whether unsupervised deep learning is appropriate,we first pose a smaller question: Can unsupervised neural net-works apply linguistic rules productively, using them in novelsituations? We draw from the literature on determiner/nounproductivity by training an unsupervised, autoencoder networkmeasuring its ability to combine nouns with determiners. Oursimple autoencoder creates combinations it has not previouslyencountered and produces a degree of overlap matching adults.While this preliminary work does not provide conclusive evi-dence for productivity, it warrants further investigation withmore complex models. Further, this work helps lay the foun-dations for future collaboration between the deep learning andcognitive science communities.

Objective

Modern healthcare data reflect massive multi-level and multi-scale information collected over many years. The majority of the existing phenotyping algorithms use case-control definitions of disease. This paper aims to study the time to disease onset and progression and identify the time-varying risk factors that drive them.

Materials and methods

We developed an algorithmic approach to phenotyping the incidence of diseases by consolidating data sources from the UK Biobank (UKB), including primary care electronic health records (EHRs). We focused on defining events, event dates, and their censoring time, including relevant terms and existing phenotypes, excluding generic, rare, or semantically distant terms, forward-mapping terminology terms, and expert review. We applied our approach to phenotyping diabetes complications, including a composite cardiovascular disease (CVD) outcome, diabetic kidney disease (DKD), and diabetic retinopathy (DR), in the UKB study.

Results

We identified 49 049 participants with diabetes. Among them, 1023 had type 1 diabetes (T1D), and 40 193 had type 2 diabetes (T2D). A total of 23 833 diabetes subjects had linked primary care records. There were 3237, 3113, and 4922 patients with CVD, DKD, and DR events, respectively. The risk prediction performance for each outcome was assessed, and our results are consistent with the prediction area under the ROC (receiver operating characteristic) curve (AUC) of standard risk prediction models using cohort studies.

Discussion and conclusion

Our publicly available pipeline and platform enable streamlined curation of incidence events, identification of time-varying risk factors underlying disease progression, and the definition of a relevant cohort for time-to-event analyses. These important steps need to be considered simultaneously to study disease progression.

Diabetes-related complications reflect longstanding damage to small and large vessels throughout the body. In addition to the duration of diabetes and poor glycemic control, genetic factors are important contributors to the variability in the development of vascular complications. Early heritability studies found strong familial clustering of both macrovascular and microvascular complications. However, they were limited by small sample sizes and large phenotypic heterogeneity, leading to less accurate estimates. We take advantage of two independent studies-UK Biobank and the Action to Control Cardiovascular Risk in Diabetes trial-to survey the single nucleotide polymorphism heritability for diabetes microvascular (diabetic kidney disease and diabetic retinopathy) and macrovascular (cardiovascular events) complications. Heritability for diabetic kidney disease was estimated at 29%. The heritability estimate for microalbuminuria ranged from 24 to 60% and was 41% for macroalbuminuria. Heritability estimates of diabetic retinopathy ranged from 6 to 33%, depending on the phenotype definition. More severe diabetes retinopathy possessed higher genetic contributions. We show, for the first time, that rare variants account for much of the heritability of diabetic retinopathy. This study suggests that a large portion of the genetic risk of diabetes complications is yet to be discovered and emphasizes the need for additional genetic studies of diabetes complications.

OBJECTIVE: To determine the relationship of metabolic weight categories with incident diabetes mellitus (DM) in postmenopausal women. METHODS: The Womens Health Initiative (WHI) enrolled 161,808 postmenopausal women aged 50 to 79 years. We included those with cardiovascular disease (CVD) biomarkers and free of CVD and prevalent DM (n = 17,043) at baseline. Normal weight was defined as a body mass index (BMI) ≥18.5 and <25 kg/m, and waist circumference (WC) <88 cm and overweight/obesity as a BMI ≥25 kg/m or WC ≥88 cm. Metabolically healthy was based on <2 and metabolically unhealthy ≥2 traits of the following: triglycerides ≥150 mg/dL, systolic blood pressure (BP) ≥130 mm Hg or diastolic BP ≥85 mm Hg, or antihypertensives or diuretics, fasting glucose ≥100 mg/dL or DM medication, and high-density lipoprotein cholesterol <50 mg/dL. Cox regression was performed to determine the risk of incident DM among metabolically healthy normal weight (MHNW), metabolically unhealthy normal weight (MUHNW), metabolically healthy overweight/obese (MHO), and metabolically unhealthy overweight/obese (MUHO). RESULTS: Among our sample, 2,253 (13.3%) participants developed DM over a mean ± standard deviation follow-up time of 15.6 ± 3.4 years. Compared with MHNW (n = 162 incident DM cases), an increased risk of incident DM was observed in MUHNW (n = 102 cases) (hazard ratio [HR] 2.24, 95% confidence interval [CI] 1.74-2.88, P < 0.0001), MHO (n = 624 cases) (HR 1.68, 95% CI 1.40-2.00, P < 0.0001), and MUHO (n = 1,365 cases) (HR 4.51, 95% CI 3.82-5.35, P < 0.0001). CONCLUSIONS: Among postmenopausal women, MUHNW and MHO confer an approximate doubling in the risk and MUHO more than a four-fold increased risk for developing DM.

Background

Obesity is associated with an increased risk of heart failure (HF); however, how metabolic weight groups relate to HF risk, especially in postmenopausal women, has not been demonstrated.

Methods

We included 19 412 postmenopausal women ages 50 to 79 without cardiovascular disease from the Women's Health Initiative. Normal weight was defined as a body mass index ≥18.5 and <25 kg/m² and waist circumference <88 cm and overweight/obesity as a body mass index ≥25 kg/m² or waist circumference ≥88 cm. Metabolically healthy was based on <2 and unhealthy ≥2 cardiometabolic traits: triglycerides ≥150 mg/dL, systolic blood pressure ≥130 mm Hg or diastolic blood pressure ≥85 mm Hg or blood pressure medication, fasting glucose ≥100 mg/dL or diabetes medication, and HDL-C (high-density lipoprotein cholesterol) <50 mg/dL. Risk factor-adjusted Cox regression examined the hazard ratios (HRs) for incident hospitalized HF among metabolically healthy normal weight (reference), metabolically unhealthy normal weight, metabolically healthy overweight/obese, and metabolically unhealthy overweight/obese.

Results

Among our sample, 455 (2.34%) participants experienced HF hospitalizations over a mean follow-up time of 11.3±1.1 years. Compared with metabolically healthy normal weight individuals, HF risk was greater in metabolically unhealthy normal weight (HR, 1.66 [95% CI, 1.01-2.72], P=0.045) and metabolically unhealthy overweight/obese individuals (HR, 1.95 [95% CI, 1.35-2.80], P=0.0004), but not metabolically healthy overweight/obese individuals (HR, 1.15 [95% CI, 0.78-1.71], P=0.48). Subdividing the overweight/obese into separate groups showed HRs for metabolically unhealthy obese of 2.62 (95% CI, 1.80-3.83; P<0.0001) and metabolically healthy obese of 1.52 (95% CI, 0.98-2.35; P=0.06).

Conclusions

Metabolically unhealthy overweight/obese and metabolically unhealthy normal weight are associated with an increased risk of HF in postmenopausal women.