Background: Predicting hypoglycemia while maintaining a low false alarm rate is a challenge for the wide adoption of continuous glucose monitoring (CGM) devices in diabetes management. One small study suggested that a deep learning model based on the long short-term memory (LSTM) network had better performance in hypoglycemia prediction than traditional machine learning algorithms in European patients with type 1 diabetes. However, given that many well-recognized deep learning models perform poorly outside the training setting, it remains unclear whether the LSTM model could be generalized to different populations or patients with other diabetes subtypes. Objective: The aim of this study was to validate LSTM hypoglycemia prediction models in more diverse populations and across a wide spectrum of patients with different subtypes of diabetes. Methods: We assembled two large data sets of patients with type 1 and type 2 diabetes. The primary data set including CGM data from 192 Chinese patients with diabetes was used to develop the LSTM, support vector machine (SVM), and random forest (RF) models for hypoglycemia prediction with a prediction horizon of 30 minutes. Hypoglycemia was categorized into mild (glucose=54-70 mg/dL) and severe (glucose<54 mg/dL) levels. The validation data set of 427 patients of European-American ancestry in the United States was used to validate the models and examine their generalizations. The predictive performance of the models was evaluated according to the sensitivity, specificity, and area under the receiver operating characteristic curve (AUC). Results: For the difficult-to-predict mild hypoglycemia events, the LSTM model consistently achieved AUC values greater than 97% in the primary data set, with a less than 3% AUC reduction in the validation data set, indicating that the model was robust and generalizable across populations. AUC values above 93% were also achieved when the LSTM model was applied to both type 1 and type 2 diabetes in the validation data set, further strengthening the generalizability of the model. Under different satisfactory levels of sensitivity for mild and severe hypoglycemia prediction, the LSTM model achieved higher specificity than the SVM and RF models, thereby reducing false alarms. Conclusions: Our results demonstrate that the LSTM model is robust for hypoglycemia prediction and is generalizable across populations or diabetes subtypes. Given its additional advantage of false-alarm reduction, the LSTM model is a strong candidate to be widely implemented in future CGM devices for hypoglycemia prediction.

Germline SDHA mutations are reported in a minority of pheochromocytoma/paraganglioma (PPGL) cases, but are associated with an increased risk of malignancy, leading some to advocate cascade genetic testing and surveillance screening of ‘at-risk’ first-degree relatives. However, such approaches rely on accurate estimates of variant pathogenicity and disease penetrance, which may have been subject to ascertainment and reporting biases, although the recent provision of large population-based DNA sequence datasets may provide a potentially unbiased resource to aid variant interpretation. Thus, the aim of the current study was to evaluate the pathogenicity and penetrance of SDHA variants reported in literature-based PPGL cases, by comparing their frequency to those occurring in the GnomAD dataset, which provides high-quality DNA sequence data on 138,632 individuals. In total, 39 different missense or loss-of-function (LOF) SDHA variants were identified in 95 PPGL index cases. Notably, many of the PPGL-associated SDHA alleles were observed at unexpectedly high frequency in the GnomAD cohort, with ∼1% and ∼0.1% of the background population harboring a rare missense or LOF variant, respectively. Although the pathogenicity of several SDHA alleles was supported by significant enrichment in PPGL cases relative to GnomAD controls, calculations of disease penetrance based on allele frequencies in the respective cohorts, resulted in much lower estimates than previously reported, ranging from 0.1-4.9%. Thus, whilst this study provides support for the etiological role of SDHA in PPGL formation, it suggests that the majority of variant carriers will not manifest PPGL and are unlikely to benefit from periodic surveillance screening.

Data on the association of CYP2C9 genetic polymorphisms with sulfonylurea ( SU )‐induced hypoglycaemia ( SH ) are inconsistent. Recent studies showed that variants in the P450 oxidoreductase ( POR ) gene could affect CYP2C9 activity. In this study, we explored the effects of POR*28 and combined CYP2C9*2 and CYP2C9*3 genotypes on SH and the efficacy of SU treatment in type 2 diabetes. A total of 1770 patients were included in the analysis of SU efficacy, assessed as the combined outcome of the HbA1c reduction and the prescribed SU daily dose. Sixty‐nine patients with severe SH were compared with 311 control patients. The number of CYP2C9 deficient alleles was associated with nearly three‐fold higher odds of hypoglycaemia ( OR , 2.81; 95% CI , 1.30‐6.09; P = .009) and better response to SU treatment (β, −0.218; SE , 0.074; P = .003) only in patients carrying the POR*1/*1 genotype. Our results indicate that interaction between CYP2C9 and POR genes may be an important determinant of efficacy and severe adverse effects of SU treatment.

Objectives The LPA single-nucleotide polymorphism rs10455872 has been associated with low-density lipoprotein cholesterol (LDLc) lowering response to statins in several randomized control trials (RCTs) and is a known coronary artery disease (CAD) marker. However, it is unclear what residual risk of CAD this marker may have during statin treatment. Methods Using electronic medical records linked to the GoDARTS genotyped population, we identified over 8000 patients on statins in Tayside, Scotland. Results We replicated the findings of the RCTs, with the G allele of rs10455872 being associated with a 0.10 mmol/l per allele poorer reduction in LDLc in response to statin treatment, and conducted a meta-analysis with previously published RCTs (P=1.46×10⁻²⁹, n=30 467). We showed an association between rs10455872 and CAD in statin-treated individuals and have replicated this finding in the Utrecht Cardiovascular Pharmacogenetics study (combined odds ratio 1.41, 95% confidence interval 1.17–1.68, P=4.5×10⁻⁵, n=8822) suggesting that statin treatment does not abrogate this well-established genetic risk for CAD. Furthermore, in a Cox proportional hazards model with LDLc measured time dependently, we demonstrated that the relationship between CAD and rs10455872 was independent of LDLc during statin treatment. Conclusion Individuals with the G allele of rs10455872, which represents approximately one in seven patients, have a higher risk of CAD than the majority of the population even after treatment with statins; and therefore represent a vulnerable group requiring an alternative medication in addition to statin treatment.