Dementia is underdiagnosed in both the general population and among Veterans. This underdiagnosis decreases quality of life, reduces opportunities for interventions, and increases health-care costs. New approaches are therefore necessary to facilitate the timely detection of dementia. This study seeks to identify cases of undiagnosed dementia by developing and validating a weakly supervised machine-learning approach that incorporates the analysis of both structured and unstructured electronic health record (EHR) data. A topic modeling approach that included latent Dirichlet allocation, stable topic extraction, and random sampling was applied to VHA EHRs. Topic features from unstructured data and features from structured data were compared between Veterans with (n = 1861) and without (n = 9305) ICD-9 dementia codes. A logistic regression model was used to develop dementia prediction scores, and manual reviews were conducted to validate the machine-learning results. A total of 853 features were identified (290 topics, 174 non-dementia ICD codes, 159 CPT codes, 59 medications, and 171 note types) for the development of logistic regression prediction scores. These scores were validated in a subset of Veterans without ICD-9 dementia codes (n = 120) by experts in dementia who performed manual record reviews and achieved a high level of inter-rater agreement. The manual reviews were used to develop a receiver of characteristic (ROC) curve with different thresholds for case detection, including a threshold of 0.061, which produced an optimal sensitivity (0.825) and specificity (0.832). Dementia is underdiagnosed, and thus, ICD codes alone cannot serve as a gold standard for diagnosis. However, this study suggests that imperfect data (e.g., ICD codes in combination with other EHR features) can serve as a silver standard to develop a risk model, apply that model to patients without dementia codes, and then select a case-detection threshold. The study is one of the first to utilize both structured and unstructured EHRs to develop risk scores for the diagnosis of dementia.

A Cardiac-centered Frailty Ontology can be an important foundation for using NLP to assess patient frailty. Frailty is an important consideration when making patient treatment decisions, particularly in older adults, those with a cardiac diagnosis, or when major surgery is a consideration. Clinicians often report patient’s frailty in progress notes and other documentation. Frailty is recorded in many different ways in patient records and many different validated frailty-measuring instruments are available, with little consistency across instruments. We specifically explored concepts relevant to decisions regarding cardiac interventions. We based our work on text found in a large corpus of clinical notes from the Department of Veterans Affairs (VA) national Electronic Health Record (EHR) database. The full ontology has 156 concepts, with 246 terms. It includes 86 concepts we expect to find in clinical documents, with 12 qualifier values. The remaining 58 concepts represent hierarchical groups (e.g., physical function findings). Our top-level class is clinical finding, which has children clinical history finding, instrument finding, and physical examination finding, reflecting the OGMS definition of clinical finding. Instrument finding is any score found for the existing frailty instruments. Within our ontology, we used SNOMED-CT concepts where possible. Some of the 86 concepts we expect to find in clinical documents are associated with the properties like ability interpretation. The concept ability to walk can either be able, assisted or unable. Each concept-property level pairing gets a different frailty score. Each scored concept received three scores: a frailty score, a relevance to cardiac decisions score, and a likelihood of resolving after the recommended intervention score. The ontology includes the relationship between scores from ten frailty instruments and frailty as assessed using ontology concepts. It also included rules for mapping ontology elements to instrument items for three common frailty assessment instruments. Ontology elements are used in two clinical NLP systems. We developed and validated a Cardiac-centered Frailty Ontology, which is a machine-interoperable description of frailty that reflects all the areas that clinicians consider when deciding which cardiac intervention will best serve the patient as well as frailty indications generally relevant to medical decisions. The ontology owl file is available on Bioportal at http://bioportal.bioontology.org/ontologies/CCFO .

The application of natural language processing and machine learning (ML) in electronic health records (EHRs) may help reduce dementia underdiagnosis, but models that are not designed to reflect minority populations may instead perpetuate underdiagnosis. To improve the identification of undiagnosed dementia, particularly in Black Americans (BAs), we developed support vector machine (SVM) ML models to assign dementia risk scores based on features identified in unstructured EHR data (via latent Dirichlet allocation and stable topic extraction in n = 1 M notes) and structured EHR data. We hypothesized that separate models would show differentiation between racial groups, so the models were fit separately for BAs (n = 5 K with dementia ICD codes, n = 5 K without) and White Americans (WAs; n = 5 K with codes, n = 5 K without). To validate our method, scores were generated for separate samples of BAs (n = 10 K) and WAs (n = 10 K) without dementia codes, and the EHRs of 1.2 K of these patients were reviewed by dementia experts. All subjects were age 65+ and drawn from the VA, which meant that the samples were disproportionately male. A strong positive relationship was observed between SVM-generated risk scores and undiagnosed dementia. BAs were more likely than WAs to have undiagnosed dementia per chart review, both overall (15.3% vs. 9.5%) and among Veterans with >90th percentile cutoff scores (25.6% vs. 15.3%). With chart reviews as the reference standard and varied cutoff scores, the BA model performed slightly better than the WA model (AUC = 0.86 with negative predictive value [NPV] = 0.98, positive predictive value [PPV] = 0.26, sensitivity = 0.61, specificity = 0.92 and accuracy = 0.91 at >90th percentile cutoff vs. AUC = 0.77 with NPV = 0.98, PPV = 0.15, sensitivity = 0.43, specificity = 0.91 and accuracy = 0.89 at >90th). Our findings suggest that race-specific ML models can help identify BAs who may have undiagnosed dementia. Future studies should examine model generalizability in settings with more females and test whether incorporating these models into clinical settings increases the referral of undiagnosed BAs to specialists.

Objectives: To investigate the relationship between serum magnesium levels, prescribed oral magnesium replacement, and major adverse cardiovascular events (MACE) in type-2 diabetes mellitus (T2D). Research design and methods: This nationwide retrospective study analyzed 1,284,940 US Veterans (≥18 years) with T2D who had outpatient serum magnesium testing between 1999–2021 in the Veterans Health Administration. The relationship between serum magnesium levels and MACE (hospitalizations for acute myocardial infarction, heart failure, ischemic stroke, or all-cause mortality) was determined using multivariable-adjusted Cox-regression models. Using a new-user-design and propensity-score-matching approach, we further related the use of prescribed oral magnesium and MACE among patients with hypomagnesemia (serum magnesium 2.3 mg/dL). Compared to patients with normomagnesemia (serum magnesium 1.8–2.3 mg/dL), those with either hypomagnesemia or hypermagnesemia had elevated hazards for MACE. The risk increased with the severity of serum magnesium abnormalities in both directions—low (hazard ratios [HRs] 1.11–1.20) and high (HRs 1.04–1.39)—in a parabolic pattern. Oral magnesium was prescribed to 9.7% and 0.7% of patients with hypomagnesemia and normomagnesemia, respectively. After propensity-score-matching balanced across 64 baseline characteristics, oral magnesium was associated with a lower MACE risk in 40,766 matched patients with hypomagnesemia (HR 0.89; 95% confidence interval [CI], 0.84–0.93), especially those on proton-pump-inhibitors or thiazides. Oral magnesium was not related to MACE in 11,838 matched patients with normomagnesemia (HR 1.07; 95% CI, 0.97–1.17). Conclusions: In patients with T2D, both hypomagnesemia and hypermagnesemia were associated with higher one-year MACE risks compared to normomagnesemia. Prescribed oral magnesium was associated with a reduced MACE risk in hypomagnesemia but not in normomagnesemia.