Predicting Diagnosis of Alzheimer's Disease and Related Dementias Using Administrative Claims

Identifying dementia from cognitive footprints in hospital records among Chinese older adults: a machine-learning study

Background

By combining theory-driven and data-driven methods, this study aimed to develop dementia predictive algorithms among Chinese older adults guided by the cognitive footprint theory.

Methods

Electronic medical records from the Clinical Data Analysis and Reporting System in Hong Kong were employed. We included patients with dementia diagnosed at 65+ between 2010 and 2018, and 1:1 matched dementia-free controls. We identified 51 features, comprising exposures to established modifiable factors and other factors before and after 65 years old. The performances of four machine learning models, including LASSO, Multilayer perceptron (MLP), XGBoost, and LightGBM, were compared with logistic regression models, for all patients and subgroups by age.

Findings

A total of 159,920 individuals (40.5% male; mean age [SD]: 83.97 [7.38]) were included. Compared with the model included established modifiable factors only (area under the curve [AUC] 0.689, 95% CI [0.684, 0.694]), the predictive accuracy substantially improved for models with all factors (0.774, [0.770, 0.778]). Machine learning and logistic regression models performed similarly, with AUC ranged between 0.773 (0.768, 0.777) for LASSO and 0.780 (0.776, 0.784) for MLP. Antipsychotics, education, antidepressants, head injury, and stroke were identified as the most important predictors in the total sample. Age-specific models identified different important features, with cardiovascular and infectious diseases becoming prominent in older ages.

Interpretation

The models showed satisfactory performances in identifying dementia. These algorithms can be used in clinical practice to assist decision making and allow timely interventions cost-effectively.

Funding

The Research Grants Council of Hong Kong under the Early Career Scheme 27110519.

Mapping morbidity 10 years prior to a diagnosis of young onset Alzheimer's disease

INTRODUCTION

Early symptoms in young onset Alzheimer's disease (YOAD) may be misinterpreted, causing delayed diagnosis. This population-based study aimed to map morbidity prior to YOAD diagnosis.

METHODS

In a register-based incidence density matched nested case-control study, we examined hospital-diagnosed morbidity for people diagnosed with YOAD in Danish memory clinics during 2016-2020 compared to controls in a 10-year period. Conditional logistic regression produced incidence rate ratios (IRRs).

RESULTS

The study included 1745 cases and 5235 controls. YOAD patients had a higher morbidity burden in the year immediately before dementia diagnosis, for certain disorders up to 10 years before. This was especially evident for psychiatric morbidity with the highest increased IRRs throughout the entire period and IRR 1.43 (95% confidence interval 1.14-1.79) in the 5-10-years before dementia diagnosis.

DISCUSSION

YOAD patients display a different pattern of morbidity up to 10 years prior to diagnosis. Awareness of specific alterations in morbidity may improve efforts toward a timely diagnosis.

HIGHLIGHTS

Retrospective, nested case-control study of young onset Alzheimer's disease (YOAD). YOAD cases had a higher morbidity burden than controls. YOAD cases had a higher psychiatric morbidity burden up to 10 years before diagnosis. Altered morbidity patterns could serve as an early warning sign of YOAD.

Artificial intelligence approaches for early detection of neurocognitive disorders among older adults

Introduction

Dementia is one of the major global health issues among the aging population, characterized clinically by a progressive decline in higher cognitive functions. This paper aims to apply various artificial intelligence (AI) approaches to detect patients with mild cognitive impairment (MCI) or dementia accurately.

Methods

Quantitative research was conducted to address the objective of this study using randomly selected 343 Saudi patients. The Chi-square test was conducted to determine the association of the patient's cognitive function with various features, including demographical and medical history. Two widely used AI algorithms, logistic regression and support vector machine (SVM), were used for detecting cognitive decline. This study also assessed patients' cognitive function based on gender and developed the predicting models for males and females separately.

Results

Fifty four percent of patients have normal cognitive function, 34% have MCI, and 12% have dementia. The prediction accuracies for all the developed models are greater than 71%, indicating good prediction capability. However, the developed SVM models performed the best, with an accuracy of 93.3% for all patients, 94.4% for males only, and 95.5% for females only. The top 10 significant predictors based on the developed SVM model are education, bedtime, taking pills for chronic pain, diabetes, stroke, gender, chronic pains, coronary artery diseases, and wake-up time.

Conclusion

The results of this study emphasize the higher accuracy and reliability of the proposed methods in cognitive decline prediction that health practitioners can use for the early detection of dementia. This research can also stipulate substantial direction and supportive intuitions for scholars to enhance their understanding of crucial research, emerging trends, and new developments in future cognitive decline studies.

The age-specific comorbidity burden of mild cognitive impairment: a US claims database study

BACKGROUND

Identifying individuals with mild cognitive impairment (MCI) who are likely to progress to Alzheimer's disease and related dementia disorders (ADRD) would facilitate the development of individualized prevention plans. We investigated the association between MCI and comorbidities of ADRD. We examined the predictive potential of these comorbidities for MCI risk determination using a machine learning algorithm.

METHODS

Using a retrospective matched case-control design, 5185 MCI and 15,555 non-MCI individuals aged ≥50 years were identified from MarketScan databases. Predictive models included ADRD comorbidities, age, and sex.

RESULTS

Associations between 25 ADRD comorbidities and MCI were significant but weakened with increasing age groups. The odds ratios (MCI vs non-MCI) in 50-64, 65-79, and ≥ 80 years, respectively, for depression (4.4, 3.1, 2.9) and stroke/transient ischemic attack (6.4, 3.0, 2.1). The predictive potential decreased with older age groups, with ROC-AUCs 0.75, 0.70, and 0.66 respectively. Certain comorbidities were age-specific predictors.

CONCLUSIONS

The comorbidity burden of MCI relative to non-MCI is age-dependent. A model based on comorbidities alone predicted an MCI diagnosis with reasonable accuracy.

Identifying Medicare beneficiaries with Alzheimer's disease and related dementia using home health OASIS assessments

BACKGROUND

Home health services are an important site of care following hospitalization among Medicare beneficiaries, providing health assessments that can be leveraged to detect diagnoses that are not available in other data sources. In this work, we aimed to develop a parsimonious and accurate algorithm using home health outcome and assessment information set (OASIS) measures to identify Medicare beneficiaries with a diagnosis of Alzheimer's disease and related dementia (ADRD).

METHODS

We conducted a retrospective cohort study of Medicare beneficiaries with a complete OASIS start of care assessment in 2014, 2016, 2018, or 2019 to determine how well the items from various versions could identify those with an ADRD diagnosis by the assessment date. The prediction model was developed iteratively, comparing the performance of different models in terms of sensitivity, specificity, and accuracy of prediction, from a multivariable logistic regression model using clinically relevant variables, to regression models with all available variables and predictive modeling techniques, to estimate the best performing parsimonious model.

RESULTS

The most important predictors of having a diagnosis of ADRD by the start of care OASIS assessment were a prior discharge diagnosis of ADRD among those admitted from an inpatient setting, and frequently exhibiting symptoms of confusion. Results from the parsimonious model were consistent across the four annual cohorts and OASIS versions with high specificity (above 96%), but poor sensitivity (below 58%). The positive predictive value was high, over 87% across study years.

CONCLUSIONS

The proposed algorithm has high accuracy, requires a single OASIS assessment, is easy to implement without sophisticated statistical models, and can be used across four OASIS versions and in situations where claims are not available to identify individuals with a diagnosis of ADRD, including the growing population of Medicare Advantage beneficiaries.

Early prediction of Alzheimer's disease and related dementias using real-world electronic health records

INTRODUCTION

This study aims to explore machine learning (ML) methods for early prediction of Alzheimer's disease (AD) and related dementias (ADRD) using the real-world electronic health records (EHRs).

METHODS

A total of 23,835 ADRD and 1,038,643 control patients were identified from the OneFlorida+ Research Consortium. Two ML methods were used to develop the prediction models. Both knowledge-driven and data-driven approaches were explored. Four computable phenotyping algorithms were tested.

RESULTS

The gradient boosting tree (GBT) models trained with the data-driven approach achieved the best area under the curve (AUC) scores of 0.939, 0.906, 0.884, and 0.854 for early prediction of ADRD 0, 1, 3, or 5 years before diagnosis, respectively. A number of important clinical and sociodemographic factors were identified.

DISCUSSION

We tested various settings and showed the predictive ability of using ML approaches for early prediction of ADRD with EHRs. The models can help identify high-risk individuals for early informed preventive or prognostic clinical decisions.

Health Conditions Associated with Alzheimer's Disease and Vascular Dementia

OBJECTIVE

We examined medical records to determine health conditions associated with dementia at varied intervals prior to dementia diagnosis in participants from the Baltimore Longitudinal Study of Aging (BLSA).

METHODS

Data were available for 347 Alzheimer's disease (AD), 76 vascular dementia (VaD), and 811 control participants without dementia. Logistic regressions were performed associating International Classification of Diseases, 9th Revision (ICD-9) health codes with dementia status across all time points, at 5 and 1 year(s) prior to dementia diagnosis, and at the year of diagnosis, controlling for age, sex, and follow-up length of the medical record.

RESULTS

In AD, the earliest and most consistent associations across all time points included depression, erectile dysfunction, gait abnormalities, hearing loss, and nervous and musculoskeletal symptoms. Cardiomegaly, urinary incontinence, non-epithelial skin cancer, and pneumonia were not significant until 1 year before dementia diagnosis. In VaD, the earliest and most consistent associations across all time points included abnormal electrocardiogram (EKG), cardiac dysrhythmias, cerebrovascular disease, non-epithelial skin cancer, depression, and hearing loss. Atrial fibrillation, occlusion of cerebral arteries, essential tremor, and abnormal reflexes were not significant until 1 year before dementia diagnosis.

INTERPRETATION

These findings suggest that some health conditions are associated with future dementia beginning at least 5 years before dementia diagnosis and are consistently seen over time, while others only reach significance closer to the date of diagnosis. These results also show that there are both shared and distinctive health conditions associated with AD and VaD. These results reinforce the need for medical intervention and treatment to lessen the impact of health comorbidities in the aging population. ANN NEUROL 2023;93:805-818.

Dementia risk predictions from German claims data using methods of machine learning

INTRODUCTION

We examined whether German claims data are suitable for dementia risk prediction, how machine learning (ML) compares to classical regression, and what the important predictors for dementia risk are.

METHODS

We analyzed data from the largest German health insurance company, including 117,895 dementia-free people age 65+. Follow-up was 10 years. Predictors were: 23 age-related diseases, 212 medical prescriptions, 87 surgery codes, as well as age and sex. Statistical methods included logistic regression (LR), gradient boosting (GBM), and random forests (RFs).

RESULTS

Discriminatory power was moderate for LR (C-statistic = 0.714; 95% confidence interval [CI] = 0.708-0.720) and GBM (C-statistic = 0.707; 95% CI  = 0.700-0.713) and lower for RF (C-statistic = 0.636; 95% CI  = 0.628-0.643). GBM had the best model calibration. We identified antipsychotic medications and cerebrovascular disease but also a less-established specific antibacterial medical prescription as important predictors.

DISCUSSION

Our models from German claims data have acceptable accuracy and may provide cost-effective decision support for early dementia screening.

External validation of existing dementia prediction models on observational health data

BACKGROUND

Many dementia prediction models have been developed, but only few have been externally validated, which hinders clinical uptake and may pose a risk if models are applied to actual patients regardless. Externally validating an existing prediction model is a difficult task, where we mostly rely on the completeness of model reporting in a published article. In this study, we aim to externally validate existing dementia prediction models. To that end, we define model reporting criteria, review published studies, and externally validate three well reported models using routinely collected health data from administrative claims and electronic health records.

METHODS

We identified dementia prediction models that were developed between 2011 and 2020 and assessed if they could be externally validated given a set of model criteria. In addition, we externally validated three of these models (Walters' Dementia Risk Score, Mehta's RxDx-Dementia Risk Index, and Nori's ADRD dementia prediction model) on a network of six observational health databases from the United States, United Kingdom, Germany and the Netherlands, including the original development databases of the models.

RESULTS

We reviewed 59 dementia prediction models. All models reported the prediction method, development database, and target and outcome definitions. Less frequently reported by these 59 prediction models were predictor definitions (52 models) including the time window in which a predictor is assessed (21 models), predictor coefficients (20 models), and the time-at-risk (42 models). The validation of the model by Walters (development c-statistic: 0.84) showed moderate transportability (0.67-0.76 c-statistic). The Mehta model (development c-statistic: 0.81) transported well to some of the external databases (0.69-0.79 c-statistic). The Nori model (development AUROC: 0.69) transported well (0.62-0.68 AUROC) but performed modestly overall. Recalibration showed improvements for the Walters and Nori models, while recalibration could not be assessed for the Mehta model due to unreported baseline hazard.

CONCLUSION

We observed that reporting is mostly insufficient to fully externally validate published dementia prediction models, and therefore, it is uncertain how well these models would work in other clinical settings. We emphasize the importance of following established guidelines for reporting clinical prediction models. We recommend that reporting should be more explicit and have external validation in mind if the model is meant to be applied in different settings.

Dementia in health claims data: The influence of different case definitions on incidence and prevalence estimates

OBJECTIVES

The epidemiology of dementia subtypes including Alzheimer's disease (AD) and vascular dementia (VD) and their reliance on different case definitions ("algorithms") in health claims data are still understudied.

METHODS

Based on health claims data, prevalence estimates (per 100 persons), incidence rates (IRs, per 100 person-years), and proportions of AD, VD, and other dementias (oD) were calculated. Five algorithms of increasing strictness considered inpatient/outpatient diagnoses (#1, #2), antidementia drugs (#3) or supportive diagnostics (#4, #5).

RESULTS

Algorithm 1 detected 213,409 cases (#2: 197,400; #3: 48,688; #4: 3033; #5: 3105), a prevalence for any dementia of 3.44 and an IR of 1.39 (AD: 0.80/0.21, VD: 0.79/0.31). The prevalence decreased by algorithms for any dementia (#2: 3.19; #3: 0.75; #4: 0.04; #5: 0.05) as did IRs (#2: 1.13; #3: 0.18; #4: 0.05, #5: 0.05). Algorithms 1-2, and 4-5 revealed similar proportions of AD (23.3%-26.6%), VD (19.9%-23.2%), and oD (53.1%-53.8%), algorithm 3 estimated 45% (AD), 12.1% (VD), and 43.0% (oD).

CONCLUSIONS

Health claims data show lower estimates of AD than previously reported, due to markedly lower prevalent/incident proportions of patients with corresponding codes. Using medication in defining dementia potentially improves estimating the proportion of AD while supportive diagnostics were of limited use.

Using dynamic microsimulation to project cognitive function in the elderly population

BACKGROUND

A long-term projection model based on nationally representative data and tracking disease progression across Alzheimer's disease continuum is important for economics evaluation of Alzheimer's disease and other dementias (ADOD) therapy.

METHODS

The Health and Retirement Study (HRS) includes an adapted version of the Telephone Interview for Cognitive Status (TICS27) to evaluate respondents' cognitive function. We developed an ordered probit transition model to predict future TICS27 score. This transition model is utilized in the Future Elderly Model (FEM), a dynamic microsimulation model of health and health-related economic outcomes for the US population. We validated the FEM TICS27 model using a five-fold cross validation approach, by comparing 10-year (2006-2016) simulated outcomes against observed HRS data.

RESULTS

In aggregate, the distribution of TICS27 scores after ten years of FEM simulation matches the HRS. FEM's assignment of cognitive/mortality status also matches those observed in HRS on the population level. At the individual level, the area under the receiver operating characteristic (AUROC) curve is 0.904 for prediction of dementia or dead with dementia in 10 years, the AUROC for predicting significant cognitive decline in two years for mild cognitive impairment patients is 0.722.

CONCLUSIONS

The FEM TICS27 model demonstrates its predictive accuracy for both two- and ten-year cognitive outcomes. Our cognition projection model is unique in its validation with an unbiased approach, resulting in a high-quality platform for assessing the burden of cognitive decline and translating the benefit of innovative therapies into long-term value to society.

Stability of individual dementia diagnoses in routine care: implications for epidemiological studies

PURPOSE

Epidemiological and health care research frequently rely on diagnoses from routine care, but the intra-individual stability of diagnoses of Alzheimer's disease (AD), vascular dementia (VD) or other forms of dementia (oD) in patients over time is understudied. More data on the diagnostic stability is needed to appraise epidemiological findings from such studies.

METHODS

Using health claims data of the years 2004-2016 from the German Pharmacoepidemiological Research Database, 160 273 patients aged ≥50 with incident dementia were identified and followed for four years. According to the incident ICD-10 codes patients were assigned to the categories AD, VD or oD. Changes between categories during follow-up were calculated.

RESULTS

Overall, 18.8% had incident AD (VD: 21.5%, oD: 59.7%). 15 842 patients had only one dementia diagnosis during four years (AD: 7.4%, VD: 12,4%, oD: 9.8%). Among those with more than one diagnosis, the incident diagnosis matched the last diagnosis in 65.1% (AD), 53.9% (VD) and 73.8% (oD) of patients. Changes in the diagnostic category were higher in patients with AD (mean: 5.1) than in patients with VD (3.6) or oD (3.3). Patients with stable AD diagnoses during the observation period were younger (median: 76 vs. 79 years) and had less inpatient treatment days (median: 14 days) than patients with changes from an AD diagnosis to another category or from another category to AD (27 days).

CONCLUSIONS

While health claims data are feasible for estimating the incidence of dementia in general, the substantial number of changes in dementia diagnoses during the course of the disease warrant caution on the interpretation of epidemiological data on specific dementia types.

Intelligent type 2 diabetes risk prediction from administrative claim data

Type 2 diabetes is a chronic, costly disease and is a serious global population health problem. Yet, the disease is well manageable and preventable if there is an early warning. This study aims to apply supervised machine learning algorithms for developing predictive models for type 2 diabetes using administrative claim data. Following guidelines from the Elixhauser Comorbidity Index, 31 variables were considered. Five supervised machine learning algorithms were used for developing type 2 diabetes prediction models. Principal component analysis was applied to rank variables' importance in predictive models. Random forest (RF) showed the highest accuracy (85.06%) among the algorithms, closely followed by the k-nearest neighbor (84.48%). The analysis further revealed RF as a high performing algorithm irrespective of data imbalance. As revealed by the principal component analysis, patient age is the most important predictor for type 2 diabetes, followed by a comorbid condition (i.e., solid tumor without metastasis). This study's finding of RF as the best performing classifier is consistent with the promise of tree-based algorithms for public data in other works. Thus, the outcome can guide in designing automated surveillance of patients at risk of forming diabetes from administrative claim information and will be useful to health regulators and insurers.

Alzheimer's Disease and Frontotemporal Dementia: A Robust Classification Method of EEG Signals and a Comparison of Validation Methods

Dementia is the clinical syndrome characterized by progressive loss of cognitive and emotional abilities to a degree severe enough to interfere with daily functioning. Alzheimer's disease (AD) is the most common neurogenerative disorder, making up 50-70% of total dementia cases. Another dementia type is frontotemporal dementia (FTD), which is associated with circumscribed degeneration of the prefrontal and anterior temporal cortex and mainly affects personality and social skills. With the rapid advancement in electroencephalogram (EEG) sensors, the EEG has become a suitable, accurate, and highly sensitive biomarker for the identification of neuronal and cognitive dynamics in most cases of dementia, such as AD and FTD, through EEG signal analysis and processing techniques. In this study, six supervised machine-learning techniques were compared on categorizing processed EEG signals of AD and FTD cases, to provide an insight for future methods on early dementia diagnosis. K-fold cross validation and leave-one-patient-out cross validation were also compared as validation methods to evaluate their performance for this classification problem. The proposed methodology accuracy scores were 78.5% for AD detection with decision trees and 86.3% for FTD detection with random forests.

Predictors of New Dementia Diagnoses in Elderly Individuals: A Retrospective Cohort Study Based on Prefecture-Wide Claims Data in Japan

Preventing dementia in elderly individuals is an important public health challenge. While early identification and modification of predictors are crucial, predictors of dementia based on routinely collected healthcare data are not fully understood. We aimed to examine potential predictors of dementia diagnosis using routinely collected claims data. In this retrospective cohort study, claims data from fiscal years 2012 (baseline) and 2016 (follow-up), recorded in an administrative claims database of the medical care system for the elderly (75 years or older) in Niigata prefecture, Japan, were used. Data on baseline characteristics including age, sex, diagnosis, and prescriptions were collected, and the relationship between subsequent new diagnoses of dementia and potential predictors was examined using multivariable logistic regression models. A total of 226,738 people without a diagnosis of dementia at baseline were followed. Of these, 26,092 incident dementia cases were detected during the study period. After adjusting for confounding factors, cerebrovascular disease (odds ratio, 1.15; 95% confidence interval, 1.11-1.18), depression (1.38; 1.31-1.44), antipsychotic use (1.40; 1.31-1.49), and hypnotic use (1.17; 1.11-1.24) were significantly associated with subsequent diagnosis of dementia. Analyses of routinely collected claims data revealed neuropsychiatric symptoms including depression, antipsychotic use, hypnotic use, and cerebrovascular disease to be predictors of new dementia diagnoses.

Targeting abnormal metabolism in Alzheimer's disease: The Drug Repurposing for Effective Alzheimer's Medicines (DREAM) study

Drug discovery for disease-modifying therapies for Alzheimer's disease and related dementias (ADRD) based on the traditional paradigm of experimental animal models has been disappointing. We describe the rationale and design of the Drug Repurposing for Effective Alzheimer's Medicines (DREAM) study, an innovative multidisciplinary alternative to traditional drug discovery. First, we use a systems biology perspective in the "hypothesis generation" phase to identify metabolic abnormalities that may either precede or interact with the accumulation of ADRD neuropathology, accelerating the expression of clinical symptoms of the disease. Second, in the "hypothesis refinement" phase we propose use of large patient cohorts to test whether drugs approved for other indications that also target metabolic drivers of ADRD pathogenesis might alter the trajectory of the disease. We emphasize key challenges in population-based pharmacoepidemiologic studies aimed at quantifying the association between medication use and ADRD onset and outline robust causal inference principles to safeguard against common pitfalls. Candidate ADRD treatments emerging from this approach will hold promise as plausible disease-modifying therapies for evaluation in randomized controlled trials.

Deep neural network models for identifying incident dementia using claims and EHR datasets

This study investigates the use of deep learning methods to improve the accuracy of a predictive model for dementia, and compares the performance to a traditional machine learning model. With sufficient accuracy the model can be deployed as a first round screening tool for clinical follow-up including neurological examination, neuropsychological testing, imaging and recruitment to clinical trials. Seven cohorts with two years of data, three to eight years prior to index date, and an incident cohort were created. Four trained models for each cohort, boosted trees, feed forward network, recurrent neural network and recurrent neural network with pre-trained weights, were constructed and their performance compared using validation and test data. The incident model had an AUC of 94.4% and F1 score of 54.1%. Eight years removed from index date the AUC and F1 scores were 80.7% and 25.6%, respectively. The results for the remaining cohorts were between these ranges. Deep learning models can result in significant improvement in performance but come at a cost in terms of run times and hardware requirements. The results of the model at index date indicate that this modeling can be effective at stratifying patients at risk of dementia. At this time, the inability to sustain this quality at longer lead times is more an issue of data availability and quality rather than one of algorithm choices.

Analysis of Benzodiazepine Prescription Practices in Elderly Appalachians with Dementia via the Appalachian Informatics Platform: Longitudinal Study

Background

Caring for the growing dementia population with complex health care needs in West Virginia has been challenging due to its large, sizably rural-dwelling geriatric population and limited resource availability.

Objective

This paper aims to illustrate the application of an informatics platform to drive dementia research and quality care through a preliminary study of benzodiazepine (BZD) prescription patterns and its effects on health care use by geriatric patients.

Methods

The Maier Institute Data Mart, which contains clinical and billing data on patients aged 65 years and older (N=98,970) seen within our clinics and hospital, was created. Relevant variables were analyzed to identify BZD prescription patterns and calculate related charges and emergency department (ED) use.

Results

Nearly one-third (4346/13,910, 31.24%) of patients with dementia received at least one BZD prescription, 20% more than those without dementia. More women than men received at least one BZD prescription. On average, patients with dementia and at least one BZD prescription sustained higher charges and visited the ED more often than those without one.

Conclusions

The Appalachian Informatics Platform has the potential to enhance dementia care and research through a deeper understanding of dementia, data enrichment, risk identification, and care gap analysis.

Decomposing Urban and Rural Disparities of Preventable ED Visits Among Patients With Alzheimer's Disease and Related Dementias: Evidence of the Availability of Health Care Resources

PURPOSE

The purpose of this study was to examine the urban and rural differences in the frequency of preventable Emergency Department (ED) visits among Alzheimer's Disease and Related Dementias (ADRD) patients, with a focus on the availability of health care resources in urban and rural areas.

METHODS

Linked datasets of 2015 State Emergency Department Databases from the Healthcare Cost and Utilization Project and the Area Health Resource File were used. ED discharges of 7 states were included in our analysis. We performed a state fixed-effect multivariable logistic regression to estimate the variation of preventable EDs by urban and rural areas. Individual characteristics and county-level health care resources were included in the estimation. The Oaxaca decomposition was used to quantify the association of county-level health care resources and urban/rural disparities.

FINDINGS

Rural patients with ADRD had 1.23 higher adjusted odds (P < .001) of going to the ED for a preventable visit compared to urban counterparts. The decomposition results showed that the model specification explained 49.2% of the differences between urban and rural patients. Patient residence in a mental health professional shortage area is one of the driving factors (contributing to 27%-48%) that explained the urban and rural disparities.

CONCLUSIONS

Our study demonstrates the importance of improving health care resources in rural areas to improve health care quality and outcomes among ADRD patients who reside in rural areas. Future research and data collection on unobserved factors, such as health care quality, will be helpful in explaining the geographic differences.

Predicting Onset of Dementia Using Clinical Notes and Machine Learning: Case-Control Study

Background

Clinical trials need efficient tools to assist in recruiting patients at risk of Alzheimer disease and related dementias (ADRD). Early detection can also assist patients with financial planning for long-term care. Clinical notes are an important, underutilized source of information in machine learning models because of the cost of collection and complexity of analysis.

Objective

This study aimed to investigate the use of deidentified clinical notes from multiple hospital systems collected over 10 years to augment retrospective machine learning models of the risk of developing ADRD.

Methods

We used 2 years of data to predict the future outcome of ADRD onset. Clinical notes are provided in a deidentified format with specific terms and sentiments. Terms in clinical notes are embedded into a 100-dimensional vector space to identify clusters of related terms and abbreviations that differ across hospital systems and individual clinicians.

Results

When using clinical notes, the area under the curve (AUC) improved from 0.85 to 0.94, and positive predictive value (PPV) increased from 45.07% (25,245/56,018) to 68.32% (14,153/20,717) in the model at disease onset. Models with clinical notes improved in both AUC and PPV in years 3-6 when notes’ volume was largest; results are mixed in years 7 and 8 with the smallest cohorts.

Conclusions

Although clinical notes helped in the short term, the presence of ADRD symptomatic terms years earlier than onset adds evidence to other studies that clinicians undercode diagnoses of ADRD. De-identified clinical notes increase the accuracy of risk models. Clinical notes collected across multiple hospital systems via natural language processing can be merged using postprocessing techniques to aid model accuracy.

Machine learning models to predict onset of dementia: A label learning approach

Introduction

The study objective was to build a machine learning model to predict incident mild cognitive impairment, Alzheimer's Disease, and related dementias from structured data using administrative and electronic health record sources.

Methods

A cohort of patients (n = 121,907) and controls (n = 5,307,045) was created for modeling using data within 2 years of patient's incident diagnosis date. Additional cohorts 3–8 years removed from index data are used for prediction. Training cohorts were matched on age, gender, index year, and utilization, and fit with a gradient boosting machine, lightGBM.

Results

Incident 2-year model quality on a held-out test set had a sensitivity of 47% and area-under-the-curve of 87%. In the 3-year model, the learned labels achieved 24% (71%), which dropped to 15% (72%) in year 8.

Discussion

The ability of the model to discriminate incident cases of dementia implies that it can be a worthwhile tool to screen patients for trial recruitment and patient management.

Generation and validation of algorithms to identify subjects with dementia using administrative data

OBJECTIVES

To generate and validate algorithms for the identification of individuals with dementia in the community setting, by the interrogation of administrative records, an inexpensive and already available source of data.

METHODS

We collected and anonymized information on demented individuals 65 years of age or older from ten general practitioners (GPs) in the district of Brianza (Northern Italy) and compared this with the administrative data of the local health protection agency (Agenzia per la Tutela della Salute). Indicators of the disease in the administrative database (diagnosis of dementia in the hospital discharge records; use of cholinesterase inhibitors/memantine; neuropsychological tests; brain CT/MRI; outpatient neurological visits) were used separately and in different combinations to generate algorithms for the detection of patients with dementia.

RESULTS

When used individually, indicators of dementia showed good specificity, but low sensitivity. By their combination, we generated different algorithms: I-therapy with ChEI/memantine or diagnosis of dementia at discharge or neuropsychological tests (specificity 97.9%, sensitivity 52.5%); II-therapy with ChEI/memantine or diagnosis of dementia at discharge or neuropsychological tests or brain CT/MRI or neurological visit (sensitivity 90.8%, specificity 70.6%); III-therapy with ChEI/memantine or diagnosis of dementia at discharge or neuropsychological tests or brain CT/MRIMRI and neurological visit (specificity 89.3%, sensitivity 73.3%).

CONCLUSIONS

These results show that algorithms obtained from administrative data are not sufficiently accurate in classifying patients with dementia, whichever combination of variables is used for the identification of the disease. Studies in large patient cohorts are needed to develop further strategies for identifying patients with dementia in the community setting.

An Algorithm to Characterize a Dementia Population by Disease Subtype

PURPOSE

Identification of Alzheimer disease and related dementias (ADRD) subtypes is important for pharmacologic treatment and care planning, yet inaccuracies in dementia diagnoses make ADRD subtypes hard to identify and characterize. The objectives of this study were to (1) develop a method to categorize ADRD cases by subtype and (2) characterize and compare the ADRD subtype populations by demographic and other characteristics.

METHODS

We identified cases of ADRD occurring during 2008 to 2014 from the OptumLabs Database using diagnosis codes and antidementia medication fills. We developed a categorization algorithm that made use of temporal sequencing of diagnoses and provider type.

RESULTS

We identified 36,838 individuals with ADRD. After application of our algorithm, the largest proportion of cases were nonspecific dementia (41.2%), followed by individuals with antidementia medication but no ADRD diagnosis (15.6%). Individuals with Alzheimer disease formed 10.2% of cases. Individuals with vascular dementia had the greatest burden of comorbid disease. Initial documentation of dementia occurred primarily in the office setting (35.1%).

DISCUSSION

Our algorithm identified 6 dementia subtypes and three additional categories representing unique diagnostic patterns in the data. Differences and similarities between groups provided support for the approach and offered unique insight into ADRD subtype characteristics.