Challenge of time series models

Transcription

1 Efficacy and efficiency of a restrictive antibiotic policy on MRSA in the intensive care unit N. Vernaz*, R. Aschbacher, B. Moser, S. Harbarth, P. Mian, P. Bonnabry, L. Pagani (Geneva, CH; Bolzano, IT) Modelling the impact of antibiotic use on antibiotic-resistant Escherichia coli using population-based data from a large hospital and its surrounding community N. Vernaz*, B. Huttner, D. Muscionico, J. Salomon, P. Bonnabry, J.M López-Lozano, J. Schrenzel, S. Harbarth (Geneva, CH; Alicante, ES) Challenge of time series models Reinforce the evidence Real-world questions Real-world data Methodological 1

2 Transfer function model ARIMA Outcomes of interest: Incidence of non-duplicate clinical isolates MRSA E. coli resistant to - ciprofloxacin - cefepime (ESBL) Explanatory variables: antibiotic usage - interventions antibiotic usage in ICU Restriction antibiotic policy (,1) antibiotic usage - surrounding community - HUG (2 beds) Intervention model Tranfer function model Efficacy and efficiency of a restrictive antibiotic policy 1) November 23: withdrawing ampicillin/sulbactam prophylaxis 2) Mai 24: targeting vancomycin therapy whenever indicated on MRSA in the intensive care unit Study period: January 22 to December 27 2

3 MRSA model identification 1 MRSA MRSA = MRSA( t-2 ) + εt DDD Restrictive antibiotic policy efficacy ampi-sulbactam, R 2 = 66% 8 22M2 22M6 22M1 23M2 23M6 23M1 24M2 24M6 24M1 25M2 25M6 25M1 26M2 26M6 Vancomycin, R 2 = 39% 26M1 27M2 27M6 27M M4 22M7 22M1 23M1 23M4 23M7 23M1 24M1 24M4 24M7 24M1 25M1 25M4 25M7 25M1 26M1 26M4 26M7 26M1 27M1 27M4 27M7 27M1 6 5 ampi-sulfa intervention ampi-sulbactam = y AR(1).379 MA(7) + ε t < November 23 y 1 November 23 DDD vancomycine intervention vancomycin = y AR(3) + ε t < Mai 24 y 1 Mai 24 3

4 Transfer function on MRSA incidence: efficiency Variable lag Coefficient Std. Error t-statistic Prob. month intercept ceftriaxone ceftriaxone levofloxacine avr AR MA Useful tool for policy maker Modelling antibiotic usage and resistance is a useful tool for antimicrobial stewardship to drive a more appropriate empirical and targeted antimicrobial therapy to control and prevent the misuse of antimicrobials to increase the level of evidence Time series analysis is a useful tool even at a ICU level 4

5 Questions? Modeling the impact of antibiotic use on antibiotic-resistant Escherichia coli using population-based data from a large hospital and its surrounding community Antibiotic prescribing quality indicator: outpatient fluoroquinolone use in 23 DDD (J1MA) x 1 DDD (J1) outside de Box plot: 18% at Geneva surrounding community excess usage of fluoroquinolones can be associated with the development of resistance ESF EMRC EXPLORATORY WORKSHOP Antwerp, Belgium, 7-9 September 25. Antibiotic Prescribing Quality Indicators. 5

6 Settings & Methods Incidence of non-duplicate clinical isolates of E. coli resistant to ciprofloxacin; of community origin (= CA -Cipro-R) ciprofloxacin; of hospital origin (= HA -Cipro-R) cefepime (= surrogate of ESBL) Antibiotic usage: Geneva surrounding community: 45 inhabitants HUG: 2 beds Defined Daily Dose (DDD) normalised per 1 patients-days per 1 inhabitants Exclude Paediatric, Psychiatry, Rehabilitation wards Study period: Jan. 2 to Dec. 27 Non-duplicate, resistant E. coli isolates.35.3 Incidence per 1 patient days R_E.Coli_ciprofloxacine R_E.Coli_cotrimoxazole ESBL 6

7 18 Total antibiotic usage surrounding community 16 DDD per 1' inhabitants janv. juin. nov. avr.1 sept.1 févr.2 juil.2 déc.2 mai.3 oct.3 mars.4 août.4 janv.5 juin.5 nov.5 avr.6 sept.6 févr.7 juil.7 déc.7 trimethoprim-sulfamethoxazole III gen. cephalosporins II gen cephalosporins macrolides fluoroquinolones amoxicillin amoxicillin/clavunate Average antimicrobial use: ( ) DDD/1 inhabitants Outpatient FQ usage Fluroquinolones consumption janv. juil. janv.1 juil.1 janv.2 juil.2 janv.3 juil.3 janv.4 juil.4 janv.5 juil.5 janv.6 juil.6 janv.7 juil.7 DID Ofloxacine Ciprofloxacine Norfloxacine Loméfloxacine Fléroxacine Lévofloxacine Moxifloxacine 7

8 Antibiotic usage Geneva Univ. Hospitals 6 DDD per 1 patient days glycopeptides trimethoprim-sulfamethoxazole macrolides carbapenems fluoroquinolones cefepime III gen cephalosporins II gen cephalosporins cefazoline amoxicillin amoxicillin/clavulanate Average antimicrobial use: ( ) DDD/1 patients-days ESBL ESBL transformation Y t -Y t-1 : D_ESBL. I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV Null Hypothesis: R_ESBL has a unit root Exogenous: None Lag Length: 4 (Automatic - based on SIC, maxlag=11) -.3 I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV Null Hypothesis: D(R_ESBL) has a unit root Exogenous: None Lag Length: 3 (Automatic - based on SIC, maxlag=11) t-statistic Prob.* t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. t*:.37 > non stationary Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. t* < stationary 8

9 Transfer function model for E. coli resistant to cefepime (ESBL) Variable Lag (months) Parameter (SE) t-statistic P-value d - ceftriaxone HUG.41 (.17) d - ciprofloxacin HUG 1.43 (.13) (.14) d - cefepime HUG 3.34 (.16) d - piperacilline/tazobactam HUG 3.99 (.42) d - ciprofloxacin GE (.8) Autoregressive term (.184) Transfer function model on ESBL incidence explains 51% of ESBL variation over time Transfer function model on E. coli resistant to ciprofloxacin, community origin ciprofloxacin, hospital origin Cipro-R- CA, R 2 =.52 Cipro-R-HA, R 2 =.18 Variable Lag Parameter t- P-value Lag Parameter t- P-value (months) (SE) Statistic (months) (SE) Statistic Constant -3.54(.24) (.1) log ciprofloxacin GE 1.31(.49) (.49) (.3) log moxifloxacin GE 4.44(.16) Autoregressive term 1.31(.11) (.1) Moving average term 8.36(.11)

10 Conclusion Added value of time series analysis: to better understand the interaction between community and hospital antibiotic prescribing Temporal relationship between: outpatient fluoroquinolone use and incidence of CA- Cipro-R- E Coli HA- Cipro-R- E Coli ESBL Support efforts to reduce prescriptions of fluoroquinolones Useful tool for policy maker: hierarchy of evidence Level I II III IV V VI Description Strong evidence from at least ONE systematic review of well designed RCT Evidence from at least one properly designed RCT of appropriate size Evidence from well designed trials without randomization: cohort study, time-series or matched case control studies Evidence from well designed non-experimental studies from more than one centre or research group Opinions from respected authorities, based on clinical evidence, descriptive studies or reports from committees Views of colleagues / peers Example Meta-Analysis The Cochrane Collaboration Articles published in peerreviewed journals Articles published in peerreviewed journals Articles published in peerreviewed journals Evidence based local procedures and care pathways Colleagues or members of the multidisciplinary team 1

11 Remember that all models are wrong; the practical question is how wrong do they have to be to not be useful G. Box, N. Draper 11