A comparison of the recording of comorbidity in primary and secondary care by using the Charlson Index to predict short-term and long-term survival in a routine linked data cohort

OBJECTIVE Hospital admission records provide snapshots of clinical histories for a subset of the population admitted to hospital. In contrast, primary care records provide continuous clinical histories for complete populations, but might lack detail about inpatient stays. Therefore, combining primary and secondary care records should improve the ability of comorbidity scores to predict survival in population-based studies, and provide better adjustment for case-mix differences when assessing mortality outcomes. DESIGN Cohort study. SETTING English primary and secondary care 1 January 2005 to 1 January 2010. PARTICIPANTS All patients 20 years and older registered to a primary care practice contributing to the linked Clinical Practice Research Datalink from England. OUTCOME The performance of the Charlson index with mortality was compared when derived from either primary or secondary care data or both. This was assessed in relation to short-term and long-term survival, age, consultation rate, and specific acute and chronic diseases. RESULTS 657,264 people were followed up from 1 January 2005. Although primary care recorded more comorbidity than secondary care, the resulting C statistics for the Charlson index remained similar: 0.86 and 0.87, respectively. Higher consultation rates and restricted age bands reduced the performance of the Charlson index, but the index's excellent performance persisted over longer follow-up; the C statistic was 0.87 over 1 year, and 0.85 over all 5 years of follow-up. The Charlson index derived from secondary care comorbidity had a greater effect than primary care comorbidity in reducing the association of upper gastrointestinal bleeding with mortality. However, they had a similar effect in reducing the association of diabetes with mortality. CONCLUSIONS These findings support the use of the Charlson index from linked data and show that secondary care comorbidity coding performed at least as well as that derived from primary care in predicting survival.