Could an integrated model of health and social care after critical illness reduce socioeconomic disparities in outcomes? A Bayesian analysis

Abstract

Background: There is limited evidence to understand what impact, if any, recovery services might have for patients across the socioeconomic spectrum after critical illness. We analysed data from a multicentre critical care recovery programme to understand the impact of this programme across the socioeconomic spectrum.

Methods: The setting for this pre-planned secondary analysis was a critical care rehabilitation programme-Intensive Care Syndrome: Promoting Independence and Return to Employment. Data were collected from five hospital sites running this programme. We utilised a Bayesian approach to analysis and explore any possible effect of the InS:PIRE intervention on Health-Related Quality of Life (HRQoL) across the socioeconomic gradient. A Bayesian quantile, non-linear mixed effects regression model, using a compound symmetry covariance structure, accounting for multiple timepoints was utilised. The Scottish Index of Multiple Deprivation (SIMD) was used to measure socioeconomic status and HRQoL was measured using the EQ-5D-5L.

Results: In the initial baseline cohort of 182 patients, 55% of patients were male, the median age was 58 yr (inter-quartile range: 50-66 yr) and 129 (79%) patients had two or more comorbidities at ICU admission. Using the neutral prior, there was an overall probability of intervention benefit of 100% (β=0.71, 95% credible interval: 0.34-1.09) over 12 months to those in the SIMD≤3 cohort, and an 98.6% (β=-1.38, 95% credible interval: -2.62 to -0.16) probability of greater benefit (i.e. a steeper increase in improvement) at 12 months in the SIMD≤3 vs SIMD≥4 cohort in the EQ-visual analogue scale.

Conclusions: Using multicentre data, this re-analysis suggests, but does not prove, that an integrated health and social care intervention is likely to improve outcomes across the socioeconomic gradient after critical illness, with a potentially greater benefit for those from deprived communities. Future research designed to prospectively analyse how critical care recovery programmes could potentially improve outcomes across the socioeconomic gradient is warranted.

Keywords: Bayesian; critical illness; deprivation; quality of life; socioeconomic.

Fig 1

Output of modelling utilising the neutal prior approach. (a) Prior-posterior density plot of EQ-5D-5L VAS trend per month. The neutral prior distribution, which followed a normal distribution with a mean of 0 and standard deviation of 0.08 (the clinically meaningful difference derived from the EQ-5D-5L VAS), is denoted in orange; the posterior distribution is shown in blue. The Bayes factor estimates the relative change between the prior and posterior distributions. (b) Prior-posterior density plot of EQ-5D-5L HUS trend per month. The neutral prior distribution, which followed a normal distribution with a mean of 0 and standard deviation of 8 (the clinically meaningful difference derived from the EQ-5D-5L HUS), is denoted in orange; the posterior distribution is shown in blue. The Bayes factor estimates the relative change between the prior and posterior distributions. (c) EQ-5D-5L VAS over time in months. β estimate of the difference in EQ-5D-5L VAS trend per month given with 95% credible intervals, and the probability of intervention benefit, Pr(β<0), indicating the likelihood of a significant difference estimate. (d) EQ-5D-5L HUS over time in months. β estimate of the difference in EQ-5D-5L HUS trend per month given with 95% credible intervals, and the probability of intervention benefit, Pr(β<0), indicating the likelihood of a significant difference estimate. CrI, credible interval; HUS, health utility score; SIMD, Scottish Index of Multiple Deprivation; VAS, visual analogue scale.