Development and application of alternative models for identifying surgical site infection: Examining the association between provider CABG volume and healthcare-associated infection
Monday, November 4, 2013
: 3:15 p.m. - 3:30 p.m.
Background Volume-infection relationships have been published for high-risk surgical procedures, although the conclusions remain controversial. Inconsistent results may be due to inaccurate infection cases identification and inconsistent classification methods of service volumes. Therefore, the purposes of this study were to develop alternative surveillance models for identifying CABG surgical site infection cases, and to examine the influence of different categorizations and the effects of long- and short-term service volumes on CABG surgical site infection (SSI). Methods Conventional ICD-9 model and classification and regression tree (CART) models were adopted to identify CABG SSI cases. Two medical centers' national health insurance claim data and healthcare-associated infection surveillance data were used for model development and validation. Patient's characteristics and medical utilization was used as parameters in CART model. In addition, the best identification model was applied to identify SSI cases in 7007 CABG cases from 19 medical centers in Taiwan from 2006-2008. Three classification methods- quartile, k-means and generalized additive model were adopted to categorize long- and short-term hospital service volumes and physician service volumes, and multilevel analysis was conducted to examine the relationship between service volumes and patient surgical site infections. Results The results showed that the length of stay, number of antibiotics, dose of cefazolin, and complexity of surgery were important factors to identified the SSI cases in CART model. The CART model had better sensitivity (87.50%), specificity (99.40%), positive predict value (77.78%) and negative predict value (99.70%) than conventional method in claim data. By CART model, 107 of 7007 CABG cases were identified as SSI cases. The results also demonstrated that the three service volume classification methods yielded different results. Categorization using the generalized additive model generated the best model fit (AIC=1083.34). The results indicate that greater cumulative physician service volume during the previous 12 months of surgery was associated with reduced risk of infection in patients (OR=0.9888). Hospitals with a medium level of service volume in the previous one month were associated with higher infection risk than hospitals with a low volume (OR=1.7510). Hospitals with a medium level of service volume in the previous 12 months were also associated with higher risk in patients than hospitals with a low volume (OR=1.9058). Conclusions The CART model has better diagnostic power for identifying SSI cases in claim data. The findings imply that the relationship between service volume and infections is not robust. An objective classification method is recommended for future research studies.
Public health or related research
Compared different models for identifying surgical site infection in Taiwan. And the best identification model was applied in Taiwan's NHI data for infection cases identification. This study also examined the relationship between providers' volume and infection cases
Keyword(s): Service Learning, Medical Care
Presenting author's disclosure statement:
Qualified on the content I am responsible for because: This was my doctoral thesis
Any relevant financial relationships? No
I agree to comply with the American Public Health Association Conflict of Interest and Commercial Support Guidelines,
and to disclose to the participants any off-label or experimental uses of a commercial product or service discussed
in my presentation.