OP4-6 - Common comorbidities and COVID-19 mortality: an empirical comparison of primary care alone vs additional hospital data linkage for association analyses

Introduction

Routinely collected primary care data in the UK has been predominantly utilised as a stand-alone resource for studying interactions between various non-communicable diseases (NCD) and COVID-19. The findings from these early studies played a crucial role in informing early policy responses to the pandemic, such as formulating the Shielded Patient List and prioritising vaccination schemes. However, few studies have been conducted to explore the potential value of adding linked hospital data to improve the completeness of recording for association studies.

Methods

A prospective cohort of 408,395 participants from UK Biobank alive on 1^st March 2020 (index date) was included, with linkage to individual-level primary care (GP) and secondary care datasets (HES, Hospital Episode Statistics). Thirteen common NCDs were studied, including hypertension, diabetes, cancer (excluding lung and haematological), haematological malignancy, asthma, chronic heart disease, chronic neurological diseases (excluding stroke and dementia), chronic respiratory disease (excluding asthma), chronic kidney disease, chronic liver disease, psychiatric disorders, common autoimmune diseases (rheumatoid arthritis, systemic lupus erythematosus and psoriasis), and dementia. We calculated the proportion of GP and HES-identified NCD cases and examined the risk of COVID-19 death associated with GP-ascertained NCD vs GP plus HES-ascertained one using multivariable Cox models.

Results

Nine out of thirteen NCDs were well captured in the GP data source, with proportions of only HES identifiable cases ranging from 8.0% for diabetes to 24.3% for chronic heart disease. However, 50.5% of dementia, 78.4% of chronic liver disease, 87.7% of chronic neurological disease and 94.2% of chronic kidney disease were not recorded in GP diagnosis records. In the subsequent association analyses, most of the studied NCDs were associated with an increased risk of COVID-19 death. The magnitude and direction of association estimates were highly comparable, either using GP or GP plus HES to ascertain NCD cases. One noticeable exception was hypertension, where analysis of GP data showed a negative association with COVID-19 death (adjusted hazard ratio 0.77, 95%CI 0.67 to 0.88), but a positive one after secondary care was added (1.34, 95%CI 1.13 to 1.59). In addition, estimation appeared to be more pronounced if based on GP-identified chronic liver disease, chronic neurological disease, and chronic kidney disease than a combination of two data sources.

Conclusion

The widely adopted assumption that primary care data can capture most chronic NCDs is generally met but shows important exceptions. Also, although the under-recording of most NCDs had a negligible impact on quantifying COVID-19 death relating to NCD, caution should be taken when a stand-alone primary care-based study produced counterintuitive results, such as a lower COVID-19 death risk associated with hypertension. In future real-world studies, evidence triangulation should be pursued by incorporating multiple data sources with varying data generation mechanisms whenever possible.