1055 SIMPLE APPROACHES TO CHARACTERISING MULTIPLE LONG-TERM CONDITIONS (MULTIMORBIDITY) AND RATES OF HOSPITAL ADMISSION IN UK BIOBANK

Introduction

Multiple approaches are used to describe MLTC, including counts and indices (weighted counts). However, few studies have directly compared different approaches which would be helpful for the design of future MLTC studies. We therefore aimed to compare the prevalence of MLTC and association with emergency hospital admission characterised using different approaches within the same dataset.

Method

We used baseline data from 501,596 UK Biobank participants (age 37-73y) combined with linked hospital episode statistics to characterise MLTC using four approaches: count of body systems affected (CB); count of 43 conditions (CC); Byles index (BI); and Charlson comorbidity index (CCI). For each we performed survival analyses to test associations to a combined endpoint of first emergency admission or death over five years from baseline interview, adjusting for age and sex. We used Harrell’s C-index to compare the discriminant value of the four approaches for the combined endpoint.

Results

Prevalence of two or more positive responses (indicating the presence of MLTC) was 45% for CB, 33% for CC, 6% for BI and 2% for CCI. 95,812 (19%) participants had one or more emergency admission or died in the first five years. All four approaches were associated with greater rates of admission/death. For example, using CC, compared to those with zero conditions those with one had 1.42 (95% CI 1.39,1.45), those with two 1.94 (95% CI 1.90,1.98), and those with 3+ conditions had 3.11 (95% CI 3.05,3.17) times greater rates. The predictive value of the four approaches was modest (C-indices ranging from 0.60 to 0.63).

Conclusion

The counts classified a greater proportion of the sample as having MLTC than the indices, highlighting that prevalence estimates of MLTC may vary depending on the approach used. All approaches had strong statistical associations with admission, but only moderate ability to identify individual participants at risk.