Using elastic nets to estimate frailty burden from routinely collected national aged care data

Machine learning analyses identify multi-modal frailty factors that selectively discriminate four cohorts in the Alzheimer's disease spectrum: a COMPASS-ND study

BACKGROUND

Frailty indicators can operate in dynamic amalgamations of disease conditions, clinical symptoms, biomarkers, medical signals, cognitive characteristics, and even health beliefs and practices. This study is the first to evaluate which, among these multiple frailty-related indicators, are important and differential predictors of clinical cohorts that represent progression along an Alzheimer's disease (AD) spectrum. We applied machine-learning technology to such indicators in order to identify the leading predictors of three AD spectrum cohorts; viz., subjective cognitive impairment (SCI), mild cognitive impairment (MCI), and AD. The common benchmark was a cohort of cognitively unimpaired (CU) older adults.

METHODS

The four cohorts were from the cross-sectional Comprehensive Assessment of Neurodegeneration and Dementia dataset. We used random forest analysis (Python 3.7) to simultaneously test the relative importance of 83 multi-modal frailty indicators in discriminating the cohorts. We performed an explainable artificial intelligence method (Tree Shapley Additive exPlanation values) for deep interpretation of prediction effects.

RESULTS

We observed strong concurrent prediction results, with clusters varying across cohorts. The SCI model demonstrated excellent prediction accuracy (AUC = 0.89). Three leading predictors were poorer quality of life ([QoL]; memory), abnormal lymphocyte count, and abnormal neutrophil count. The MCI model demonstrated a similarly high AUC (0.88). Five leading predictors were poorer QoL (memory, leisure), male sex, abnormal lymphocyte count, and poorer self-rated eyesight. The AD model demonstrated outstanding prediction accuracy (AUC = 0.98). Ten leading predictors were poorer QoL (memory), reduced olfaction, male sex, increased dependence in activities of daily living (n = 6), and poorer visual contrast.

CONCLUSIONS

Both convergent and cohort-specific frailty factors discriminated the AD spectrum cohorts. Convergence was observed as all cohorts were marked by lower quality of life (memory), supporting recent research and clinical attention to subjective experiences of memory aging and their potentially broad ramifications. Diversity was displayed in that, of the 14 leading predictors extracted across models, 11 were selectively sensitive to one cohort. A morbidity intensity trend was indicated by an increasing number and diversity of predictors corresponding to clinical severity, especially in AD. Knowledge of differential deficit predictors across AD clinical cohorts may promote precision interventions.

Determinants of Higher Mortality at Six Months in Patients with Hip Fracture: A Retrospective Study

(1) Background: Hip fracture is a pathology with high mortality, but the lack of a universal adaptation of the factors associated with death makes it difficult to predict risk and implement prevention in this group. This study aimed to identify the factors that determine a higher mortality at six months following hip fracture. (2) Methods: A retrospective longitudinal study, whose study population consisted of patients over 65 years of age. The main variable was mortality at 6 months of fracture. Relevant data related to sociodemographic and clinical variables for subsequent bivariate (χ²) and multivariate analysis were obtained. (3) Results: In all, 665 people participated in the study, 128 of whom died within 6 months of the fracture. The multivariate adjusted analysis demonstrated significant relationships between the main variable and aspects such as institutionalization at discharge (Odds Ratio (OR) = 2.501), a worse overall functional capacity (OR = 2.453) and cognitive capacity (OR = 3.040) at admission, and complications such as heart failure (OR = 5.767) or respiratory infection (OR = 5.308), in addition to the taking of certain drugs and the presence of a greater number of comorbidities. (4) Conclusions: There are certain factors related to higher mortality at six months in patients with hip fracture who are aged 65 years or older.

The risk of fall-related hospitalisations at entry into permanent residential aged care

Background

Entering permanent residential aged care (PRAC) is a vulnerable time for individuals. While falls risk assessment tools exist, these have not leveraged routinely collected and integrated information from the Australian aged and health care sectors. Our study examined individual, system, medication, and health care related factors at PRAC entry that are predictors of fall-related hospitalisations and developed a risk assessment tool using integrated aged and health care data.

Methods

A retrospective cohort study was conducted on N = 32,316 individuals ≥65 years old who entered a PRAC facility (01/01/2009-31/12/2016). Fall-related hospitalisations within 90 or 365 days were the outcomes of interest. Individual, system, medication, and health care-related factors were examined as predictors. Risk prediction models were developed using elastic nets penalised regression and Fine and Gray models. Area under the receiver operating characteristics curve (AUC) assessed model discrimination.

Results

64.2% (N = 20,757) of the cohort were women and the median age was 85 years old (interquartile range 80-89). After PRAC entry, 3.7% (N = 1209) had a fall-related hospitalisation within 90 days and 9.8% (N = 3156) within 365 days. Twenty variables contributed to fall-related hospitalisation prediction within 90 days and the strongest predictors included fracture history (sub-distribution hazard ratio (sHR) = 1.87, 95% confidence interval (CI) 1.63-2.15), falls history (sHR = 1.41, 95%CI 1.21-2.15), and dementia (sHR = 1.39, 95%CI 1.22-1.57). Twenty-seven predictors of fall-related hospitalisation within 365 days were identified, the strongest predictors included dementia (sHR = 1.36, 95%CI 1.24-1.50), history of falls (sHR = 1.30, 95%CI 1.20-1.42) and fractures (sHR = 1.28, 95%CI 1.15-1.41). The risk prediction models had an AUC of 0.71 (95%CI 0.68-0.74) for fall-related hospitalisations within 90 days and 0.64 (95%CI 0.62-0.67) for within 365 days.

Conclusion

Routinely collected aged and health care data, when integrated at a clear point of action such as entry into PRAC, can identify residents at risk of fall-related hospitalisations, providing an opportunity for better targeting risk mitigation strategies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12877-021-02640-w.

Predictors of short-term hospitalization and emergency department presentations in aged care

OBJECTIVES

To examine individual, medication, system, and healthcare related predictors of hospitalization and emergency department (ED) presentation within 90 days of entering the aged care sector, and to create risk-profiles associated with these outcomes.

DESIGN AND SETTING

Retrospective population-based cohort study using data from the Registry of Senior Australians.

PARTICIPANTS

Older people (aged 65 and older) with an aged care eligibility assessment in South Australia between January 1, 2013 and May 31, 2016 (N = 22,130).

MEASUREMENTS

Primary outcomes were unplanned hospitalization and ED presentation within 90 days of assessment. Individual, medication, system, and healthcare related predictors of the outcomes at the time of assessment, within 90 days or 1-year prior. Fine-Gray models were used to calculate subdistribution hazard ratios (sHR) and 95% confidence intervals (CI). Harrell's C-index assessed predictive ability.

RESULTS

Four thousand nine-hundred and six (22.2%) individuals were hospitalized and 5028 (22.7%) had an ED presentation within 90 days. Predictors of hospitalization included: being a man (hospitalization sHR = 1.33, 95% CI 1.26-1.42), ≥3 urgent after-hours attendances (hospitalization sHR = 1.21, 95% CI 1.06-1.39), increasing frailty index score (hospitalization sHR = 1.19, 95% CI 1.11-1.28), individuals using glucocorticoids (hospitalization sHR = 1.11, 95% CI 1.02-1.20), sulfonamides (hospitalization sHR = 1.18, 95% CI 1.10-1.27), trimethoprim antibiotics (hospitalization sHR = 1.15, 95% CI 1.03-1.29), unplanned hospitalizations 30 days prior (hospitalization sHR = 1.13, 95% CI 1.04-1.23), and ED presentations 1 year prior (hospitalization sHR = 1.07, 95% CI 1.04-1.10). Similar predictors and hazard estimates were also observed for ED presentations. The hospitalization models out-of-sample predictive ability (C-index = 0.653, 95% CI 0.635-0.670) and ED presentations (C-index = 0.647, 95% CI 0.630-0.663) were moderate.

CONCLUSIONS

One in five individuals with aged care eligibility assessments had unplanned hospitalizations and/or ED presentation within 90 days with several predictors identified at the time of aged care eligibility assessment. This is an actionable period for targeting at-risk individuals to reduce hospitalizations.

Health status and healthcare trends of individuals accessing Australian aged care programmes over a decade: the Registry of Senior Australians historical cohort

Background

Understanding the health profile, service and medicine use of Australians in the aged care sector will help inform appropriate service provision for our ageing population.

Aims

To examine the 2006–2015 trends in (i) comorbidities and frailty of individuals accessing aged care, and (ii) health services, medicine use and mortality after entry into long‐term care.

Methods

Cross‐sectional and population‐based trend analyses were conducted using the Registry of Senior Australians.

Results

From 2006 to 2015, 509 944 individuals accessed permanent residential care, 206 394 home care, 283 014 respite and 124 943 transition care. Over this time, the proportion of individuals accessing permanent residential care with high frailty scores (≥0.3) increased (19.7–49.7%), as did the proportion with 5–9 comorbidities (46.4–54.5%), with similar trends observed for those accessing other services. The median number of medicines dispensed in the year after entering permanent residential care increased from 9 (interquartile range (IQR) 6–12) to 10 (IQR 7–14), while remaining stable in home care (2006: 9, IQR 5–12, 2015: 9, IQR 6–13). Short‐term (within 100 days) mortality in those accessing permanent care was higher in 2006 (15.6%, 95% CI 15.2–16.0) than 2015 (14.6%, 95% CI 14.3–14.9). Longer term (101–1095 days, 2006: 44.3%, 95% CI 43.7–45.0, 2015: 46.4%, 95% CI 45.8–46.9) mortality was higher in 2015 compared to 2006. Mortality in individuals accessing home care did not change.

Conclusion

The health of older Australians accessing aged care programmes has declined while frailty increased, with an increasing use of medicine and worse long‐term mortality in some. Funding and care models need to adapt to this changing profile.

Utilisation of general practice health assessments around an aged care assessment is associated with lower mortality risk in older Australians

OBJECTIVE

(i) to describe the general practitioner utilisation of health assessments, management plans, coordination of team care arrangements and medication review item numbers within 6 months of an aged care eligibility assessment for home care packages (HCP) and (ii) investigate the impact of health assessments on the risk of mortality and entry into permanent residential aged care (PRAC) of individuals accessing HCP.

DESIGN AND SETTING

retrospective cohort study utilising data from the Registry of Senior Australians (ROSA) was conducted.

SUBJECTS

75,172 individuals aged ≥75 years who received HCP between 2011 and 2015.

OUTCOME MEASURE

for objective 1: the use of comprehensive assessments (Medicare Benefits Schedule (MBS) items 705 or 707), management plans (MBS 721), coordination of team care arrangements (MBS 723), and medication reviews (MBS 900). For objective 2: time to death and entry into PRAC.

RESULTS

of the 75,172 individuals, 28.2% (95% confidence interval (CI): 27.8-8.5%) had comprehensive assessments, 36.7% (95% CI: 36.3-37.0%) had management plans, 33.0% (95% CI: 32.7-33.3%) received coordination of team care arrangements and 5.4% (95% CI: 5.2-5.5%) had medication reviews. Individuals with a comprehensive assessment had a 5% lower risk of mortality (adjusted hazard ratio (aHR), 95% CI = 0.95, 0.92-0.98) but 5% higher risk of transition to PRAC (adjusted subdistribution HRs, 95% CI = 1.05, 1.02-1.08) compared to those who did not have these services.

CONCLUSION

the utilisation of health assessments was associated with a lower risk of mortality. There is an opportunity for increased use of item numbers in frailer individuals.

The effect of frailty on outcomes of surgically treated hip fractures in older people

BACKGROUND

Hip fractures are associated with mortality, disability, and loss of independence in older adults. While several risk factors associated with poor outcomes following a hip fracture have been identified, the effect of frailty status prior to hip fracture is not well established.

AIM

To examine the associations of frailty with mortality, change in activities of daily living (ADL) limitations, and transition to permanent residential aged care in older people following a hip fracture.

METHODS

A retrospective cohort study was conducted on people aged 65 years and older with a surgically treated hip fracture between 2003 and 2015. Frailty was estimated using a cumulative deficit-based frailty index and categorized into quartiles. Cox multivariable regression, logistic regression, and Fine-Gray multivariable regression models estimated associations of frailty with mortality, ADL limitations, and entry into permanent residential aged care, respectively. Hazard ratios (HR), odds ratios (OR), subdistribution hazard ratios (SHR), and 95% confidence intervals (95%CI) are reported.

RESULTS

Out of 4771 individuals with hip fractures, 75.6% were female and the median age was 86 (interquartile range 82-90) years old. The two-year survival of patients following hip fracture was 43.7% (95%CI 40.9-46.7%) in those in the highest quartile of frailty, compared to 54.4% (95%CI 51.8-57.2%) for those in the lowest quartile (HR = 1.25, 95%CI 1.11-1.41, p < 0.001). No associations between pre-fracture frailty and post-fracture ADL limitations were observed. Additionally, no association of frailty with transition to permanent residential aged care for patients living in the community (n = 1361) was observed (SHR = 0.98, 95%CI 0.81-1.18, p = 1.000).

CONCLUSIONS

Older patients with the highest level of frailty had an increased risk of mortality after hip fracture. Consideration for appropriate clinical interventions, including fall and frailty prevention measures, may be appropriate for this identified group of vulnerable individuals.