1088. Prediction Model to Identify Patients at Risk of 30-day Treatment Failure in Patients with Staphylococcus aureus Bacteremia
Session: Poster Abstract Session: Clinical Infectious Diseases: Bacteremia and Endocarditis
Friday, October 28, 2016
Room: Poster Hall

Background: Clinical or microbiological treatment failure and mortality from Staphylococcus aureus bacteremia (SAB) remains significant.  A predictive model was developed to identify patients at increased risk of treatment failure in order to improve management strategies. Methods: Adults with monomicrobial SAB from two prospective observational cohorts were included.  Treatment failure was a composite of all-cause mortality, persistent bacteremia (³ 7 days) and recurrent bacteremia within 30 days.  Candidate variables were selected a priori based on clinical significance and prior literature, and a parsimonious model was derived using the Akaike Information Criteria with a complete case analysis.  A sum score was derived for clinical use. Results: Complete data were available for 388 patients, and 20.6% experienced 30-day treatment failure.  The parsimonious multivariable model identified severe sepsis, methicillin-resistant S. aureus (MRSA), patient age, and serum albumin as predictors of treatment failure.  Model calibration (Hosmer-Lemeshow test, p=0.376) and discrimination (c-statistic 0.752) indicated good performance.  A prediction sum score was developed from this model (Table 1); Figure 1 shows the predicted probability of treatment failure and Figure 2 shows the sum score according to observed outcomes.  The sum score performed well (c-statistic 0.750 compared with the multivariable model). Conclusion: The combination of severe sepsis, MRSA, older patient age and lower serum albumin were predictors of 30-day treatment failure.  Although some of these factors are non-modifiable, these are useful indicators for the clinician to ensure the patient receives early appropriate antibiotics and aggressive supportive care to improve clinical outcomes.

Table 1.  Sum score from parsimonious model to predict risk of 30-day treatment failure.


Model coefficient

Rounded coefficient

Score assigned*

Severe sepsis








Age (per 10 year increment from age 18 years)




Albumin (per 5 g/L increment from 10g/L)




* Possible range of score: -7 to +13

Figure 1.  Predicted probability of 30-day treatment failure using sum score derived from parsimonious predictive model.

Figure 2.  Performance of predictive model sum score versus observed 30-day treatment outcome.

Natasha Holmes, MBBS FRACP PhD, Department of Infectious Diseases, Austin Health, Heidelberg VIC, Australia, J Owen Robinson, MD FRACP, Department of Microbiology and Infectious Diseases, Royal Perth Hospital & Fiona Stanley Hospital, Perth WA, Australia, Sebastian Van Hal, MBBS FRACP FRCPA PhD, Infectious Diseases and Microbiology, Royal Prince Alfred Hospital, Sydney, Australia, Wendy Munckhof, MBBS FRACP FRCPA PhD, Infection Management Services, Princess Alexandra Hospital, Woolloongabba QLD, Australia, Eugene Athan, MBBS, FRACP, MPH, Infectious Diseases, Barwon Health, Deakin University, Geelong, Australia, Tony Korman, MBBS FRACP FRCPA, Department of Infectious Diseases, Monash Infectious Diseases, Clayton VIC, Australia, Allen C. Cheng, MBBS, FRACP, MPH, PhD, Infection Prevention and Hospital Epidemiology Unit, The Alfred Hospital, Melbourne, Australia, Matthew O'sullivan, MBBS MMed FRACP FRCPA PhD, Centre for Infectious Diseases and Microbiology, Westmead Hospital, Westmead NSW, Australia, Tara Anderson, MBBS FRACP, Department of Infectious Diseases,, Royal Hobart Hospital, Hobart TAS, Australia, Sally a. Roberts, MBChB, FRACP, FRCPA, Labplus, Department of Microbiology, Auckland District Health Board, Auckland, New Zealand, Sanchia Warren, MBBS FRACP, Department of Infectious Diseases, Royal Hobart Hospital, Hobart TAS, Australia, John Turnidge, MBBS FRACP FRCPA, Australian Commission on Safety and Quality in Health Care, Sydney NSW, Australia, Paul Johnson, MBBS FRACP PhD, Infectious Diseases, Austin Health/University of Melbourne, Melbourne, Australia, Benjamin Howden, MBBS FRACP FRCPA PhD, Microbiological Diagnostic Unit, Parkville VIC, Australia and VANESSA study group


N. Holmes, National Health and Medical Research Council: Grant Investigator , Grant recipient

J. O. Robinson, None

S. Van Hal, None

W. Munckhof, None

E. Athan, None

T. Korman, None

A. C. Cheng, None

M. O'sullivan, None

T. Anderson, None

S. A. Roberts, None

S. Warren, None

J. Turnidge, None

P. Johnson, None

B. Howden, National Health and Medical Research Council: Grant Investigator , Grant recipient

Findings in the abstracts are embargoed until 12:01 a.m. CDT, Wednesday Oct. 26th with the exception of research findings presented at the IDWeek press conferences.