OWE-018 Generating better information to inform services for alcohol-related liver disease: the connected health cities programme

    loading  Checking for direct PDF access through Ovid

Abstract

Introduction

The rising burden and high mortality for alcohol-related liver disease (ARLD) is well-recognised in the UK. Efforts to plan adequate liver and alcohol services and to identify unwarranted variation are limited by lack of robust, actionable information about true workload and real-world risk-adjusted outcomes. We report an informatics programme (Connected Health Cities, Northwest Coast) that is generating new analytical resources to support services.

Methods

A healthcare data lab. team of data scientists is linked to a secure ‘data ark’ hosting regional commissioning datasets (Admitted Patient Care, A and E, Outpatients, fiscal years 14/15 to 16/17) and is working with clinical teams to construct novel ways to interrogate data (‘algorithmic approach’), model outcomes and unexplained variation and visualise data (small area mapping). Entire patient journeys were linked, to identify ARLD cases and their all-cause emergency admissions (categorised by diagnosis), capture phenotyping flags (e.g. first occurrence of codes for varices), pre-admission events and outcomes (e.g. in-hospital mortality; all-cause 30 day readmission). Maps were used to communicate catchment areas and identify admission ‘hot spots’. Site-level benchmarking reports were shared with teams at 7 hospitals.

Results

Compared to the standard approach for capturing cases (i.e. primary diagnosis at discharge; 6 specific ICD-10 codes; ‘ARLD-Primary’ Method), the algorithmic approach (ARLD-Alg) identified 9 other patterns of primary and secondary diagnoses with >60 primary ICD-10 codes consistent with emergency care for ARLD. Activity: Across 7 NHS Trusts, estimates of total 3 year activity for ARLD increased from 3185 (ARLD-Primary) to 5912 (ARLD-Alg) for admissions, 35 840 to 56 010 for bed days (equivalent to 50 extra in-patient beds), and 485 to 665 for in-hospital deaths. Case-mix for the 2 approaches was similar (mean age: 51.6 v 51.4; males 63.2 v 63.6%; Charlson Index: 1.6 v 1.7) but ARLD-Alg captured more shorter stays (median LoS: 7 v 6 days;% short stay [0–2 days]: 12.1 v 16.5). Varices cases account for 15%–20%.

Outcomes

Crude mortality across hospitals varied significantly by method ARLD-Primary: 13.0%–21.7%; ARLD-Alg: 10.2%–14.5%. All binary logistic models identified male patients were at lower risk of in-hospital death than females (Adj OR: 0.70). Treatment provider was a predictor of outcome.

Conclusions

Standard approaches to analysis of administrative data under-estimate the burden of unplanned care for ARLD (by up to 60%) and have dubious value for service planning or benchmarking. More sophisticated clinically-driven Methods can leverage greater value from routine data. These novel tools are scalable for nationwide deployment and have potential to inform policy and support data-driven service improvement.

Funding

Department of Health

Related Topics

    loading  Loading Related Articles