(AP) Longitudinal Progression of Frailty in Older Population and Risk of Mortality: An Application of Joint Latent Class Models

Publication Number: 658

Background: A validated electronic frailty index (eFI) based on the presence of 36 cumulative deficits in the electronic primary care record has been implemented in the UK to better target clinical care. It is assumed that frailty is a progressive condition that takes years to develop, and that there are subgroups of the population that may deteriorate faster than the rest.

Objectives: To evaluate the application of joint latent class models (JLCM) in identifying trajectories of frailty progression over time and risk of death in older people using routinely collected data.

Methods: Patients identified from the UK Clinical Practice Research Datalink GOLD database linked to Hospital Episode Statistics and Office for National Statistics mortality register aged ≥65 on December 31st 2009 were included. eFI scores were calculated at baseline (2009) and annually in subsequent years (2010-2013). JLCM include the longitudinal and time-to-event sub-models. It uses latent class modelling to divide the population into subgroups (clusters) with different average intercepts and slopes. The model then calculates the mortality risk of each cluster. Two- to six- cluster models were fitted. To determine the best model (i.e., number of clusters), Bayesian Inference Criterion (BIC) was used. Models were adjusted for baseline age and gender. Intercepts and slopes, baseline characteristics, and mortality hazard ratios (HR) of the identified clusters were reported.

Results: The 5-cluster model was chosen based on BIC and 475,503 patients were included. The frail-rapid cluster (n=9,104 (1.9%)) had the highest intercept (intercept 7.9, slope 1.0), followed by frail-slow cluster (n=45,707 (9.6%), (intercept 7.4, slope 0.3)), while fit-rapid cluster (n=12,869 (2.7%)) had the steepest slope (intercept 2.4, slope 1.7). Two clusters were fit with slow and moderate progression: (n=338,029 (71.1%)) and (n=69,794 (14.7%)), respectively. Fit-rapid and frail-rapid clusters had older population (Mean 79.1 (SD 7.2)) and (78.2 (7.1)), respectively, compared to other clusters (Range 74.3 to 77.0 (7.2 to 7.5)). Taking frail-rapid cluster as the reference, fit-rapid cluster had a comparable HR; 1.03 (95% CI 1.02 to 1.04), while all other clusters had lower risk of mortality (HR < 1). Polypharmacy and hypertension were among the highest prevalent deficits across all clusters and timepoints.

Conclusions: Our research showed that there are vulnerable subgroups of the older population who are currently frail or have rapid frailty progression and need more healthcare monitoring and resources. Both, current value and slope of eFI over time are associated with the risk of death. Further research into the generalisability of the resultant clusters is needed.