STUDY AIM/s:
In line with the proposed topic (3.3.4) of CIBERFes on “”Advanced data analysis and management systems (Big Data solutions) in the identification and assessment of frailty risk profiles in populations, their evolution to disability, and their treatment””), we propose the following four aims:
1.ASSESSMENT OF FRAILTY RISK PROFILES:
1.1.To validate the Rosen, Dubois, ERA and eFI frailty indices in our population by testing their validity to predict key features of the frailty phenotype (Rockwood et al [ref]) including: co-morbidity, cognitive impairment, disability, and geriatric syndromes (falls, delirium, and/or urinary incontinence) in our population aged >75 years.
1.2.To characterise the descriptive epidemiology of frailty as defined by the previously validated indices (1.1) by estimating the point prevalence and annual incidence rate of frailty after stratification by age (5-year bands), gender, and socio-economic status (quintiles of national MEDEA deprivation score).
2.EVOLUTION FROM FRAILTY TO DISABILITY
2.1.To study the impact of frailty (as defined by previously validated indices above) on the risk of subsequent severe falls (leading to GP contact/s and/or hospital admission), fracture/s, and all-cause mortality in an elderly population aged >75 years.
AND 3.TREATMENT OF FRAILTY
3.1.To study the association between oral bisphosphonate use and the risk of fracture/s and all-cause mortality amongst patients previously identified as frail (as defined above) amongst elderly people aged >75 years.
SUMMARY OF METHODOLOGY:
-Study Design: Population-based cohort study using de-identified primary care records linked to pharmacy dispensations and hospital admissions data for >80% of the population of Catalonia (SIDIAP database).
-Data Source/s: SIDIAP (Sistema de Informacion para el Desarrollo de la Investigacion en Atencion Primaria) contains de-identified information on socio-demographics (year of birth, gender, country of origin, socio-economic status, etc), lifestyle risk factors (smoking, alcohol drinking, physical activity, etc), clinical measurements (height, weight, waist circumference, blood pressure, etc), co-morbidity (recorded diagnoses using ICD-10 codes), prescriptions, and linked community pharmacy dispensations (as coded using the World Health Organization ATC catalogue). SIDIAP holds data on a total of 884,687 people aged >75 in the period 2007-2014, and contributing staff (nurses and primary care physicians) have ?as part of routine practice- collected information on activities of daily living (Barthel) and cognitive function (Pfeiffer test) for a substantial 354,457 (40%), and 346,576 (39%) of them respectively. This, linked to hospital admissions data, constitutes a unique dataset for the study of frailty, its impact, and its possible treatment in community-dwelling elderly patients.
-Participant/s: all patients aged >75 and registered in SIDIAP in the study period (2007-2015) will be eligible for Aims 1.1, 1.2 and 2.1. Previous users (before categorized as ?frail?) of anti-osteoporosis medications (except calcium/D supplementation) will be excluded for Aim 3.1.
-Study Outcome/s: the three proposed frailty indices will be validated against the Rockwood frailty criteria, which will therefore be study outcomes for Aim 1.1, to include: co-morbidity (as measured using the Charlson co-morbidity index), cognitive impairment (Pfeiffer score), disability (Barthel), and geriatric syndromes (falls, delirium, and/or urinary incontinence, all three identified using ICD10 codes in medical records).
Outcomes for Aims 2.1 and 3.1 will include severe fall/s, fracture/s (both identified using ICD10 codes), and all-cause mortality (as based on linked administrative data).
-Exposure/s: the proposed frailty indices will be the main exposure/s for Aims 1.1 and 2.1. The Rosen index will be estimated using coded diagnoses (ICD10) in SIDIAP records. Similarly, prescriptions and pharmacy dispensations will be used for the estimation of the Dubois index. Finally, the ERA index will be estimated using socio-demographics (age, gender), previous hospital admissions (from linked hospital data), and past medical history (ICD10 codes in primary care records) recorded in SIDIAP for the study population. Bisphosphonate use will be the exposure of interest for Aim 3.1. Patients will be identified as bisphosphonate users if they collected at least 2 packs of any oral bisphosphonate (ATC codes M05BA except M05BA03 and M05BA08 as these are intravenous and therefore not available in pharmacy dispensation/s data) in the study period.
-Co-variables/Confounders: lists of a priori-defined confounders will be established for each of the study Aims based on previous literature and clinical knowledge within research team members. A propensity score for bisphosphonate use will be calculated to minimize confounding in Aim 3.1.
-Sample size/power: According to feasibility figures provided by SIDIAP, at least 346,576 patients will be available for the validation of frailty indices (Aim 1.1), which requires data on Barthel/Pfeiffer scores. The whole available population aged >75 years (N=884,687) will be included for Aims 1.2 and 2.1, providing very precise estimates of validity (1.1) and prevalence/incidence (1.2). Assuming a 10% prevalence of frailty, and with known 5% annual mortality and 2/1,000 person-years fracture rates in SIDIAP patients of this age [Pages-Castella et al. BMC MSK Disord], these figures provide >99% for the detection of an excess risk of >20% (HR 1.2) as significant in the analysis of the association between frailty and fractures, with even higher power for mortality [Aim 2.1].
According to SIDIAP feasibility numbers, 87,254 bisphosphonate users and 797,433 non-users will be eligible for Aim 3.1. These numbers provide >99% power for the detection of a >5% risk reduction in mortality and >95% for a >15% fracture risk reduction associated with bisphosphonate use in a log-rank two-sided test with assumed 10% drop-outs and up to 8 years of follow-up (drop-out/transfer out was 8.2% in the period 2007-2014).
-Statistical Analyses:
Aim 1.1: Sensitivity, specificity, predictive values (negative/positive) and area under the ROC (receiver operating characteristic) curves will be provided for the validation of each of the frailty indices against each of the Rockwood criteria. Indices with an AUC ROC>70% in their validation against disability, cognitive impairment and at least one geriatric syndrome (see above) will be considered valid for the assessment of frailty.
Aim 1.2: Point (on last day before data extraction) prevalence and average annual incidence rates (for the whole study period) of frailty, as estimated using previously validated index/indices, will be estimated for the whole eligible population (aged >75) and then after stratification by age, gender, and socio-economic status
Aim 2.1: Cox regression and Fine and Gray (accounting for a competing risk with death) models will then be fitted to estimate relative risk of fall/s, fracture/s, and all-cause death according to frailty status
Aim 3.1: Propensity scores will be estimated for bisphosphonate use using logistic regression equations, and propensity score matching methods will be applied to obtain ?comparable? populations of bisphosphonate users and matched non-users. After that, similar survival methods (Cox and Fine and Gray) will be used to estimate the relative risk of fracture/s and death according to bisphosphonate use in the propensity-matched population.
The assumption of hazards proportionality will be checked for all the fitted Cox/Fine and Gray models by graphical visualization of hazard function/s and/or formal testing using Schoenfeld?s residuals.