Using large language model (LLM) to identify high-burden informal caregivers in long-term care.

BACKGROUND: The rising global elderly population increases the demand for caregiving, yet traditional methods may not fully assess the challenges faced by vital informal caregivers. OBJECTIVE: To investigate the efficacy of Large Language Model (LLM) in detecting overburdened informal caregivers, benchmarking against rule-based and machine learning methods. METHODS: 1,791 eligible informal caregivers from Southern Taiwan and utilized their textual case summary reports for the LLM. We also employed structured questionnaire results for machine learning models. Furthermore, we leveraged the visualization of the LLM's attention mechanisms to enhance our understanding of the model's interpretative capabilities. RESULTS: The LLM achieved an Area Under the Receiver Operating Characteristic (AUROC) curve of 0.84 and an Area Under the Precision-Recall Curve (AUPRC) of 0.70, marking an 8% and 14% improvement over traditional methods. The visualization of the attention mechanism accurately reflected the evaluations of human experts, concentrating on descriptions of high-burden descriptions and the relationships between caregivers and recipients. CONCLUSION: This research demonstrates the notable capability of LLM to accurately identify high-burden caregivers in Long-term Care (LTC) settings. Compared to traditional approaches, LLM offers an opportunity for the future of LTC research and policymaking.

Understanding and alleviating informal caregiver burden through the development and validation of a caregiver strain index-based model in Taiwan.

BACKGROUND: Quantifying the informal caregiver burden is important for understanding the risk factors associated with caregiver overload and for evaluating the effectiveness of services provided in Long-term Care (LTC). OBJECTIVE: This study aimed to develop and validate a Caregiver Strain Index (CSI)-based score for quantifying the informal caregiver burden, while the original dataset did not fully cover evaluation items commonly included in international assessments. Subsequently, we utilized the CSI-based score to pinpoint key caregiver burden risk factors, examine the initial timing of LTC services adoption, and assess the impact of LTC services on reducing caregiver burden. METHODS: The study analyzed over 28,000 LTC cases in Southern Taiwan from August 2019 to December 2022. Through multiple regression analysis, we identified significant risk factors associated with caregiver burden and examined changes in this burden after utilizing various services. Survival analysis was employed to explore the relationship between adopting the first LTC services and varying levels of caregiver burden. RESULTS: We identified 126 significant risk factors for caregiver burden. The most critical factors included caregiving for other disabled family members or children under the age of three (ß = 0.74, p < 0.001), the employment status of the caregiver (ß = 0.30-0.53, p < 0.001), the frailty of the care recipient (ß = 0.28-0.31, p < 0.001), and the behavioral symptoms of dementia in care recipients (ß = 0.28-2.60, p < 0.05). Generally, caregivers facing higher burdens sought LTC services earlier, and providing home care services alleviated the caregiver's burden. CONCLUSION: This comprehensive study suggests policy refinements to recognize high-risk caregivers better early and provide timely support to improve the overall well-being of both informal caregivers and care recipients.

Machine learning models for predicting blood pressure phenotypes by combining multiple polygenic risk scores.

We construct non-linear machine learning (ML) prediction models for systolic and diastolic blood pressure (SBP, DBP) using demographic and clinical variables and polygenic risk scores (PRSs). We developed a two-model ensemble, consisting of a baseline model, where prediction is based on demographic and clinical variables only, and a genetic model, where we also include PRSs. We evaluate the use of a linear versus a non-linear model at both the baseline and the genetic model levels and assess the improvement in performance when incorporating multiple PRSs. We report the ensemble model's performance as percentage variance explained (PVE) on a held-out test dataset. A non-linear baseline model improved the PVEs from 28.1 to 30.1% (SBP) and 14.3% to 17.4% (DBP) compared with a linear baseline model. Including seven PRSs in the genetic model computed based on the largest available GWAS of SBP/DBP improved the genetic model PVE from 4.8 to 5.1% (SBP) and 4.7 to 5% (DBP) compared to using a single PRS. Adding additional 14 PRSs computed based on two independent GWASs further increased the genetic model PVE to 6.3% (SBP) and 5.7% (DBP). PVE differed across self-reported race/ethnicity groups, with primarily all non-White groups benefitting from the inclusion of additional PRSs. In summary, non-linear ML models improves BP prediction in models incorporating diverse populations.

Pulse pressure is associated with decline in physical function in older adults.

OBJECTIVES: This study examined the associations between pulse pressure, hypertension, and the decline in physical function in a prospective framework. STUDY DESIGN: The Healthy Aging Longitudinal Study tracked a group of Taiwanese adults aged 55 or more over an average of 6.19 years to assess pulse pressure and decline in physical function, including in handgrip strength, gait speed, and 6-min walking distance, at baseline (2009-2013) and in the second phase of assessments (2013-2020). MAIN OUTCOME MEASURES: Pulse pressure was calculated as the difference between systolic and diastolic blood pressure values. Weakness, slowness, and low endurance were defined as decreases of ≥0.23 m/s (one standard deviation) in gait speed, ≥5.08 kg in handgrip strength, and ≥ 57.73 m in a 6-min walk, as determined from baseline to the second phase of assessment. Linear and logistic regressions were employed to evaluate the associations between pulse pressure, hypertension, and decline in physical function. RESULTS: Baseline pulse pressure was associated with future handgrip strength (beta = -0.017, p = 0.0362), gait speed (beta = -0.001, p < 0.0001), and 6-min walking distance (beta = -0.470, p < 0001). In multivariable models, only handgrip strength (beta = -0.016, p = 0.0135) and walking speed (beta = -0.001, p = 0.0042) remained significantly associated with future pulse pressure. Older adults with high systolic blood pressure (≥140 mmHg) and elevated pulse pressure (≥60 mmHg) exhibited a significantly increased risk of weakness (odds ratio: 1.30, 95 % confidence interval: 1.08-1.58), slowness (1.29, 1.04-1.59), and diminished endurance (1.25, 1.04-1.50) compared with the reference group, who exhibited systolic blood pressure of <140 mmHg and pulse pressure of <60 mmHg. CONCLUSIONS: Among older adults, pulse pressure is associated with a decline in physical function, especially in terms of strength and locomotion.

Machine learning models for blood pressure phenotypes combining multiple polygenic risk scores.

We construct non-linear machine learning (ML) prediction models for systolic and diastolic blood pressure (SBP, DBP) using demographic and clinical variables and polygenic risk scores (PRSs). We developed a two-model ensemble, consisting of a baseline model, where prediction is based on demographic and clinical variables only, and a genetic model, where we also include PRSs. We evaluate the use of a linear versus a non-linear model at both the baseline and the genetic model levels and assess the improvement in performance when incorporating multiple PRSs. We report the ensemble model's performance as percentage variance explained (PVE) on a held-out test dataset. A non-linear baseline model improved the PVEs from 28.1% to 30.1% (SBP) and 14.3% to 17.4% (DBP) compared with a linear baseline model. Including seven PRSs in the genetic model computed based on the largest available GWAS of SBP/DBP improved the genetic model PVE from 4.8% to 5.1% (SBP) and 4.7% to 5% (DBP) compared to using a single PRS. Adding additional 14 PRSs computed based on two independent GWASs further increased the genetic model PVE to 6.3% (SBP) and 5.7% (DBP). PVE differed across self-reported race/ethnicity groups, with primarily all non-White groups benefitting from the inclusion of additional PRSs.