Assessing harm in gastrointestinal surgery using linked routine national databases

Abstract

Background: Hospital administrative data have been used to evaluate population-level surgical outcome in the UK. However, these data do not account for harm recorded outside of the hospital setting, nor pertain to health status as health status is under-represented by conventional diagnostic coding. A systematic review of the literature identified studies reporting on changes in quality of life associated with surgical adverse events (SAEs). The evaluated studies demonstrated a significant correlation between short-term surgical harm and long-term physical and mental well-being. Interactions between harm and health status have not previously been studied using population-level data. Hypothesis: Patient-level linkage of multiple routine databases can improve our understanding, measurement and risk adjustment of harm in gastrointestinal surgery. Methods: Almost a tenth of primary care patients in England with linkage of Clinical Practice Research Datalink, Hospital Episodes Statistics, National Cancer Intelligence Network, Office of National Statistics and Index of Multiple Deprivation databases were studied. Nine gastrointestinal surgical procedures were evaluated to measure short-term harm (technical SAEs, systemic SAEs) and long-term health status (postoperative psychiatric morbidity, postoperative symptoms). Risk factors derived from linked databases were evaluated against outcome including 30-day mortality, readmission, prolonged hospital stay, technical SAEs, systemic SAEs, one-year mortality, postoperative psychiatric morbidity and postoperative symptoms by binary logistic regression analysis. Predictive performance was compared between models with predictor covariates derived from linked databases and those derived from hospital data only by assessing discrimination. Results: Overall, 73655 patients who underwent 75854 procedures between April 2000 and March 2011 were evaluated. At least one SAE within 30 days of surgery was recorded in 10.7% of procedures. Of all recorded SAEs, 28% were identified in primary care data only. The proportion of post-discharge SAEs was 42.5%. Within twelve months after surgery, postoperative psychiatric morbidity was recorded in 11.4% of procedures and postoperative symptoms in 16.9% of procedures. Regression analysis revealed that technical SAEs were associated with increased risk of poor postoperative health status. When adjusted for SAEs, new onset postoperative psychiatric morbidity was associated with significantly increased risk of one-year mortality (OR=1.48, p<0.001). Postoperative mental health status therefore had a significant impact on survival, and this association was particularly strong in patients undergoing surgery for upper gastrointestinal malignancy. Prospective studies must therefore explore the role of early recognition and development of interventions to reduce the impact of this type of harm. Linked databases improved the performance of prediction models for all outcome. However, this was though to be of clinical value in models predicting postoperative psychiatric morbidity only. Conclusions: Linked routine databases characterized the overall burden of surgical harm and poor health status in gastrointestinal surgery. Using this important information resource, interactions between short and long-term outcome were identified. Prediction of poor health status improved to a level that could inform policy. Further development of analytical methods and validation of data may pave the way for large-scale evaluations of quality within integrated health systems using linked routine databases.