OBJECTIVE: To develop and validate a deep learning model using readily available clinical information to predict treatment success with the first ASM for individual patients.
DESIGN, SETTING, AND PARTICIPANTS: This cohort study developed and validated a prognostic model. Patients were treated between 1982 and 2020. All patients were followed up for a minimum of 1 year or until failure of the first ASM. A total of 2404 adults with epilepsy newly treated at specialist clinics in Scotland, Malaysia, Australia, and China between 1982 and 2020 were considered for inclusion, of whom 606 (25.2%) were excluded from the final cohort because of missing information in 1 or more variables.
EXPOSURES: One of 7 antiseizure medications.
MAIN OUTCOMES AND MEASURES: With the use of the transformer model architecture on 16 clinical factors and ASM information, this cohort study first pooled all cohorts for model training and testing. The model was trained again using the largest cohort and externally validated on the other 4 cohorts. The area under the receiver operating characteristic curve (AUROC), weighted balanced accuracy, sensitivity, and specificity of the model were all assessed for predicting treatment success based on the optimal probability cutoff. Treatment success was defined as complete seizure freedom for the first year of treatment while taking the first ASM. Performance of the transformer model was compared with other machine learning models.
RESULTS: The final pooled cohort included 1798 adults (54.5% female; median age, 34 years [IQR, 24-50 years]). The transformer model that was trained using the pooled cohort had an AUROC of 0.65 (95% CI, 0.63-0.67) and a weighted balanced accuracy of 0.62 (95% CI, 0.60-0.64) on the test set. The model that was trained using the largest cohort only had AUROCs ranging from 0.52 to 0.60 and a weighted balanced accuracy ranging from 0.51 to 0.62 in the external validation cohorts. Number of pretreatment seizures, presence of psychiatric disorders, electroencephalography, and brain imaging findings were the most important clinical variables for predicted outcomes in both models. The transformer model that was developed using the pooled cohort outperformed 2 of the 5 other models tested in terms of AUROC.
CONCLUSIONS AND RELEVANCE: In this cohort study, a deep learning model showed the feasibility of personalized prediction of response to ASMs based on clinical information. With improvement of performance, such as by incorporating genetic and imaging data, this model may potentially assist clinicians in selecting the right drug at the first trial.
METHODS: This randomized, open-label, phase 3 study was conducted at 36 medical centers in China (mainland), Malaysia, South Korea, and Taiwan. Patients were randomly assigned 1:1 to 200 mg of pembrolizumab intravenously every 3 weeks for ≤2 years or 80 mg/m2 of paclitaxel intravenously every week. Primary end points were overall survival (OS) and progression-free survival (PFS). Secondary end points were objective response rate (ORR) per Response Evaluation Criteria in Solid Tumors version 1.1 and safety.
RESULTS: Between February 16, 2017, and March 12, 2018, 94 patients were randomly assigned (47 pembrolizumab/47 paclitaxel) after screening; enrollment was stopped on March 12, 2018, based on the results of the global KEYNOTE-061 study, and patients were followed until the last patient's last visit. Median OS was 8 months (95% confidence interval [CI], 4-10 months) with pembrolizumab versus 8 months (95% CI, 5-11 months) with paclitaxel (hazard ratio [HR], 0.99; 95% CI, 0.63-1.54). Median PFS was 2 months (95% CI, 1-3 months) with pembrolizumab versus 4 months (95% CI, 3-6 months) with paclitaxel (HR, 1.62; 95% CI, 1.04-2.52). ORR was 13% for pembrolizumab versus 19% for paclitaxel. Any-grade treatment-related adverse events occurred in 28 pembrolizumab-treated patients (60%) and 42 paclitaxel-treated patients (96%); grades 3 to 5 events occurred in 5 patients (11%) and 28 patients (64%), respectively.
CONCLUSIONS: Definitive conclusions about the efficacy of second-line pembrolizumab in Asian patients with advanced PD-L1-positive gastric/GEJ cancer are limited because of insufficient power, but pembrolizumab was well tolerated in this patient population. Efficacy followed a trend similar to that observed in the phase 3 KEYNOTE-061 trial.
METHODS: The development data set comprised 138,309 women from 17 case-control studies. PRSs were generated using a clumping and thresholding method, lasso penalized regression, an Empirical Bayes approach, a Bayesian polygenic prediction approach, or linear combinations of multiple PRSs. These PRSs were evaluated in 89,898 women from 3 prospective studies (1592 incident cases).
RESULTS: The best performing PRS (genome-wide set of single-nucleotide variations [formerly single-nucleotide polymorphism]) had a hazard ratio per unit SD of 1.62 (95% CI = 1.46-1.80) and an area under the receiver operating curve of 0.635 (95% CI = 0.622-0.649). Combined Asian and European PRSs (333 single-nucleotide variations) had a hazard ratio per SD of 1.53 (95% CI = 1.37-1.71) and an area under the receiver operating curve of 0.621 (95% CI = 0.608-0.635). The distribution of the latter PRS was different across ethnic subgroups, confirming the importance of population-specific calibration for valid estimation of breast cancer risk.
CONCLUSION: PRSs developed in this study, from association data from multiple ancestries, can enhance risk stratification for women of Asian ancestry.
METHODS: We performed whole genome sequencing and analyzed 6 199 696 common variants among 113 aromatic ASM-induced SJS/TEN cases and 84 tolerant controls of Han Chinese ethnicity.
RESULTS: In the primary analysis, nine variants reached genome-wide significance (p
RESEARCH DESIGN AND METHODS: The China Kadoorie Biobank recruited 512,891 adults (59% women) aged 30-79 from 10 regions of China during 2004-2008. At baseline survey, and subsequent resurveys of a random subset of survivors, participants were interviewed and measurements collected, including on-site RPG testing. Cause of death was ascertained via linkage to local mortality registries. Cox regression yielded adjusted HR for all-cause and cause-specific mortality associated with usual levels of RPG.
RESULTS: During median 11 years' follow-up, 37,214 deaths occurred among 452,993 participants without prior diagnosed diabetes or other chronic diseases. There were positive log-linear relationships between RPG and all-cause, cardiovascular disease (CVD) (n=14,209) and chronic kidney disease (CKD) (n=432) mortality down to usual RPG levels of at least 5.1 mmol/L. At RPG <11.1 mmol/L, each 1.0 mmol/L higher usual RPG was associated with adjusted HRs of 1.14 (95% CI 1.12 to 1.16), 1.16 (1.12 to 1.19) and 1.44 (1.22 to 1.70) for all-cause, CVD and CKD mortality, respectively. Usual RPG was positively associated with chronic liver disease (n=547; 1.45 (1.26 to 1.66)) and cancer (n=12,680; 1.12 (1.09 to 1.16)) mortality, but with comparably lower risks at baseline RPG ≥11.1 mmol/L. These associations persisted after excluding participants who developed diabetes during follow-up.
CONCLUSIONS: Among Chinese adults without diabetes, higher RPG levels were associated with higher mortality risks from several major diseases, with no evidence of apparent thresholds below the cut-points for diabetes diagnosis.
OBJECTIVE: To investigate the genetics of depression among individuals of East Asian and European descent living in different geographic locations, and with different outcome definitions for depression.
DESIGN, SETTING, AND PARTICIPANTS: Genome-wide association analyses followed by meta-analysis, which included data from 9 cohort and case-control data sets comprising individuals with depression and control individuals of East Asian descent. This study was conducted between January 2019 and May 2021.
EXPOSURES: Associations of genetic variants with depression risk were assessed using generalized linear mixed models and logistic regression. The results were combined across studies using fixed-effects meta-analyses. These were subsequently also meta-analyzed with the largest published GWAS for depression among individuals of European descent. Additional meta-analyses were carried out separately by outcome definition (clinical depression vs symptom-based depression) and region (East Asian countries vs Western countries) for East Asian ancestry cohorts.
MAIN OUTCOMES AND MEASURES: Depression status was defined based on health records and self-report questionnaires.
RESULTS: There were a total of 194 548 study participants (approximate mean age, 51.3 years; 62.8% women). Participants included 15 771 individuals with depression and 178 777 control individuals of East Asian descent. Five novel associations were identified, including 1 in the meta-analysis for broad depression among those of East Asian descent: rs4656484 (β = -0.018, SE = 0.003, P = 4.43x10-8) at 1q24.1. Another locus at 7p21.2 was associated in a meta-analysis restricted to geographically East Asian studies (β = 0.028, SE = 0.005, P = 6.48x10-9 for rs10240457). The lead variants of these 2 novel loci were not associated with depression risk in European ancestry cohorts (β = -0.003, SE = 0.005, P = .53 for rs4656484 and β = -0.005, SE = 0.004, P = .28 for rs10240457). Only 11% of depression loci previously identified in individuals of European descent reached nominal significance levels in the individuals of East Asian descent. The transancestry genetic correlation between cohorts of East Asian and European descent for clinical depression was r = 0.413 (SE = 0.159). Clinical depression risk was negatively genetically correlated with body mass index in individuals of East Asian descent (r = -0.212, SE = 0.084), contrary to findings for individuals of European descent.
CONCLUSIONS AND RELEVANCE: These results support caution against generalizing findings about depression risk factors across populations and highlight the need to increase the ancestral and geographic diversity of samples with consistent phenotyping.