- Email: [email protected]
Electronic measures of surgical site infection: Implications for estimating risk and costs
S100
Surgical Forum Abstracts
J Am Coll Surg
(2005-2006) and testing (2007, 2008). This was consistently higher than the AUROC for work RVU (Table). CPT SVM (Mortality AUROC)
RVU (Mortality AUROC)
2005-2006
0.82
0.657
2007
0.80
2008
0.80
Year
p-Value
CPT SVM (Morbidity AUROC)
RVU (Morbidity AUROC)
p-Value
⬍0.001
0.76
0.685
⬍0.001
0.687
⬍0.001
0.75
0.701
⬍0.001
0.683
⬍0.001
0.75
0.693
⬍0.001
Table: Comparison of SVM models mapping CPT to perioperative risk and RVU. SVM model trained on 2005-2006 PUF, and tested on 2007 PUF and 2008 PUF.
CONCLUSIONS: Non-parametric methods from artificial intelligence can translate CPT codes to perioperative risk. This approach is fully-automated and may significantly improve upon the use of work RVU in multivariable risk models like NSQIP.
Electronic measures of surgical site infection: Implications for estimating risk and costs Melissa M Boltz DO, Christopher S Hollenbeak PhD, Lucas E Nikkel BA, Eric Schaefer MS, Gail Ortenzi RN, BSN, Peter W Dillon MD Penn State Milton S. Hershey Medical Center, Hershey, PA INTRODUCTION: Electronic measures of surgical site infections (SSI) are being used more frequently in place of more labor intensive measures. This study compares the performance characteristics of two electronic measures of SSI to a traditional measure and the
implications of using electronic measures to estimate risk factors and costs attributed to SSI among general and vascular surgery patients. METHODS: Data included 1,066 patients undergoing general and vascular surgery procedures at a single academic center between 2007-2008. Clinical data were from our institution’s National Surgical Quality Improvement Program (NSQIP) database, which includes SSI as a key outcome. We compared this measure of infection to the MedMined Nosocomial Infection Marker (NIM) and International Classification of Disease, Ninth Revision (ICD-9) codes from billing records. We compared rates of infection according to each measure, estimated sensitivity and specificity of the electronic measures, compared the estimated effect of SSI measure on risk factors for mortality using logistic regression, and compared the estimated attributable cost of SSI according to each measure using linear regression. RESULTS: SSI was observed in 8.8% of patients with the NSQIP definition, 2.6% of patients with the NIM definition, and 5.8% for the ICD-9 definition. Logistic regression for each SSI measure revealed large differences in estimated risk factors. The NIM and ICD-9 measures overestimated attributable cost of SSI by 134% and 33%, respectively. CONCLUSIONS: In this data set, electronic measures of SSI did not correlate well with more traditional measures, and their use had serious implications for estimating risk factors and costs.
Surgical Forum Abstracts
J Am Coll Surg
(2005-2006) and testing (2007, 2008). This was consistently higher than the AUROC for work RVU (Table). CPT SVM (Mortality AUROC)
RVU (Mortality AUROC)
2005-2006
0.82
0.657
2007
0.80
2008
0.80
Year
p-Value
CPT SVM (Morbidity AUROC)
RVU (Morbidity AUROC)
p-Value
⬍0.001
0.76
0.685
⬍0.001
0.687
⬍0.001
0.75
0.701
⬍0.001
0.683
⬍0.001
0.75
0.693
⬍0.001
Table: Comparison of SVM models mapping CPT to perioperative risk and RVU. SVM model trained on 2005-2006 PUF, and tested on 2007 PUF and 2008 PUF.
CONCLUSIONS: Non-parametric methods from artificial intelligence can translate CPT codes to perioperative risk. This approach is fully-automated and may significantly improve upon the use of work RVU in multivariable risk models like NSQIP.
Electronic measures of surgical site infection: Implications for estimating risk and costs Melissa M Boltz DO, Christopher S Hollenbeak PhD, Lucas E Nikkel BA, Eric Schaefer MS, Gail Ortenzi RN, BSN, Peter W Dillon MD Penn State Milton S. Hershey Medical Center, Hershey, PA INTRODUCTION: Electronic measures of surgical site infections (SSI) are being used more frequently in place of more labor intensive measures. This study compares the performance characteristics of two electronic measures of SSI to a traditional measure and the
implications of using electronic measures to estimate risk factors and costs attributed to SSI among general and vascular surgery patients. METHODS: Data included 1,066 patients undergoing general and vascular surgery procedures at a single academic center between 2007-2008. Clinical data were from our institution’s National Surgical Quality Improvement Program (NSQIP) database, which includes SSI as a key outcome. We compared this measure of infection to the MedMined Nosocomial Infection Marker (NIM) and International Classification of Disease, Ninth Revision (ICD-9) codes from billing records. We compared rates of infection according to each measure, estimated sensitivity and specificity of the electronic measures, compared the estimated effect of SSI measure on risk factors for mortality using logistic regression, and compared the estimated attributable cost of SSI according to each measure using linear regression. RESULTS: SSI was observed in 8.8% of patients with the NSQIP definition, 2.6% of patients with the NIM definition, and 5.8% for the ICD-9 definition. Logistic regression for each SSI measure revealed large differences in estimated risk factors. The NIM and ICD-9 measures overestimated attributable cost of SSI by 134% and 33%, respectively. CONCLUSIONS: In this data set, electronic measures of SSI did not correlate well with more traditional measures, and their use had serious implications for estimating risk factors and costs.