Overview on determination of ECOFF, PK/PD breakpoints ,clinical cut-offs and the setting of Clinical breakpoint by VetCAST VMAS symposium/workshop Uppsala 12-15 December 2017 Pierre-Louis Toutain, Royal Veterinary College, London & National Veterinary School of Toulouse, France

Objectives of the presentation To explain how VetCAST expect to establish veterinary clinical breakpoints As a subcommittee of EUCAST, VetCAST will follow the general procedure implemented by EUCAST Based on its SOP 1.2 2016, Position paper “En route towards European Clinical Breakpoint for veterinary antimicrobial susceptibility testing: the VetCAST approach”

1-What is VetCAST

VetCAST is a Subcommittee of EUCAST

Acronyms EUCAST CLSI CLSI/VAST European Committee on Antimicrobial Susceptibility Testing CLSI The Clinical and Laboratory Standards Institute CLSI/VAST Sub-committee of CLSI dealing with the susceptibility testing of veterinary pathogens and determination of veterinary CBP

2-What is the mission of VetCAST

To establish CBP ECOFF COCL PK/PD BP

Definitions Cut-off Clinical breakpoints (CBP) the values of MIC (mg/L) selected by an ad hoc committee to be used by testing laboratories to qualitatively report the results of AST as Susceptible (S), Intermediate (I), or Resistant (R). Not to be confused with cut-offs Cut-off A threshold value for MICs to separate two entities or targets (ECOFF, PK/PD cut-off, Clinical cut-off)

Definitions ECOFF Clinical cut-off Epidemiological (bacteriological) cut-off PK/PD breakpoint (EUCAST) or PK/PD cut-off (CLSI) the MIC value for which a targeted PK/PD index value can be achieved with a routine dosage regimen in a given percentage of animals (usually 90%) Clinical cut-off Term not used by EUCAST; for CLSI, Clinical cut-off is the MIC value for which probability-of-cure can be achieved with a routine dosage regimen in a given percentage of animals

Terminology: EUCAST vs. CLSI/VAST Organisation Epidemio PK/PD Clinics Decision CLSI/VAST COWT COPD COCL CBP EUCAST ECOFF PK/PD breakpoint VETCAST PK/PD cutoff

VETCAST modus operandi: A risk analysis approach assessment Risk management Risk communication Cut-offs Clinical breakpoint Expert rules Science Decision Harmonisation

Development of a (clinical) Breakpoint (BP) Diameter Clinical BP COWT ECOFFs COPK/PD COCL

3-Establishing ECOFFs Epidemiological/microbiological cut-off

Epidemiological cut-off values (ECOFFs) ECOFFs are related to the distribution of MICs of wild type organisms lacking acquired or mutational resistance to the antimicrobial agent in question. The ECOFF is essentially the upper MIC value of the wild type distribution. EUCAST SOP 7.0

MIC-data available tetra/doxycycline EUCAST

ECOFF at VetCAST A minimal total number of 100 MIC values in the putative wild type population for each bacterial species, originating from at least 5 accepted MIC data sets are required to define an ECOFF Stakeholders are encouraged to send their in-house data for aggregation, after review by VetCAST, with the already published MICs

Methods for defining the wild type distribution (determining the ECOFF) The “eyeball” method (Kahlmeter) The 95% rule (Pfaller) The Normalised Resistance Interpretation (Kronvall) The iterative statistical method (Turnidge) Multimodal analysis (Meletiadis) From John Turnidge, CLSI Workshop, Tampa, January 2013

Ciprofloxacin & Acinetobacter baumannii

Statistical method for ECOFF/COWT

For more details

ECOFF at VetCAST A CBP will never lower than the ECOFF (see rule 10.2 of the SOP 1.2 (European Committee on Antimicrobial Susceptibility Testing, 2013) As a surrogate of CBP Local administrations, when several CBP are required … For surveillance ECOFF is a parameter, not CPB

3-The setting of a PK/PD breakpoint

The human PK/PD breakpoint reflect clinical data AUC/MIC or T>MIC magnitude predictive of cure

The PK/PD breakpoint at VetCast For veterinary medicine in the EU, reference to PK/PD concepts was only introduced in 2016 by EMA/CVMP in its latest guidance on the demonstration of efficacy of antimicrobial substances, although only for preclinical investigation (European Medicines Agency, 2016). No clinical data regarding PK/PD as for human medicine PK/PD will only reflect preclinical data and is equivalent to the PK/PD cut-off of the CLSI

The setting of a PK/PD BP Step1 Selection of a PK/PD index predictive of clinical efficacy and/or prevention of resistance Step 2 Determination of the critical value (magnitude) of the selected PK/PD index Step 3 Computation , for a given animal species and for all possible (not probable) MICs of the percentage (proportion) of animals able to achieve the critical value of the selected PK/PD index (computation of so-called Probability Target Attainment (or PTA)

Step 1:Selection of a PK/PD index fAUC24h/MIC fCmax/MIC fT>MIC fCmax/MIC was historically seen with aminoglycosides but is now losing favour, aminoglycosides being now managed as an AUC/MIC drug class , Rem: it was shown that for all drugs/formulations having a long half-live that AUC/MIC is the appropriate index (AU/MIC: universal index?)

Rodents + clinical validation Step 1: Determination of the best PK/PD index and of its critical value (magnitude ) PK/PD index Default value Experimental value Preclinical In vitro/In vivo Dynamic: killing curves In vivo Rodents + clinical validation Clinical

Dynamic: killing curves Step 1: Determination of the best PK/PD index and of its critical value (magnitude ) PK/PD index Default value Experimental value Preclinical In vitro/In vivo Dynamic: killing curves In vivo Target species Clinical

Dynamic: killing curves Step 1: Determination of the best PK/PD index and of its critical value (magnitude ) PK/PD index Default value Experimental value Preclinical In vitro/In vivo Dynamic: killing curves In vivo Target species Clinical

PK/PD indices critical values: Killing curves & hollow fibers test systems

Modeling approach of killing curve to determine PK/PD index PD parameters Florfenicol EC50 (mg/L) 0.447 Emax (1/h) 2.01 Hill 2.7

In silico dose fractionation trials Solving the semimechanistic PD model with PK parameters corresponding to a given dosage regimen and a given formulation, the time-development of the bacterial population can be predicted (in silico clinical trial)

Benzylpenicillin; ceforuxime; erythromycin; gentamicin; moxifloxacin and vancomycin The in silico predictions based on the in vitro PKPD confirmed previously established PK/PD indices The selection and magnitude of the PK/PD index were shown to be sensitive to differences in PK and dosing intervals. This study supports the use of PKPD models built from in vitro time-kill curves in the development of optimal dosing regimens for antibacterial drugs.

AUC/MIC vs. Time>MIC to predict florfenicol effect (in silico simulations) AUC/MIC=113-136h for 120h T>MIC=43-49% ( 120h) AUC/MIC T>MIC

The case of fT>MIC For a given dose, T>MIC is dependent of the shape of the curve influence of modalities of administration E.g. dosage interval Top feeding vs. pump Influence of the formulation

Short (Black) vs. LA formulations (Blue) (same AUC and T>MIC=50%) Are they equivalent in terms of T>MIC?

Short vs. LA formulations Curves T>MIC (%) AUC Black 50 5000 Blue 100 10000 Magenta 6250 -----MIC 2 options: two different CBP for Blue and Magenta or alternatively to consider the AUC s

AUC/MIC: an universal PK/PD index for veterinary medicine? It is quoted: For drugs like the b-lactams where the efficacy generally have been found to be correlated to T.MIC, the best PK/PD index shifts toward AUC/MIC dependence as the half-life increase, as seen in patients with a reduced renal function e.g., elderly or neonates.

The setting of a PK/PD CO Step1 Selection of a PK/PD index predictive of clinical efficacy and/or prevention of resistance Step 2 Determination of the critical value (size) of the selected PK/PD index Step 3 Computation , for a given animal species and for all possible (not probable) MICs of the percentage (proportion) of animals able to achieve the critical value of the selected PK/PD index (computation of so-called Probability of Target Attainment (PTA)

What is PTA? More literally the question is: E.g. What is the critical (maximal) MIC for which we can guarantee that plasma drug concentration will be: above a possible MIC for X% of the dosage interval (often 50 or 80%) for fT>MIC or≥ to a given fAUC/MIC for the intended dosage regimen in at least 90% of subjects of the targeted population

PTA (Quantile) 90 with or without its 90% Confidence interval

What PTA target value at VetCAST? Statistically, the PK/PD breakpoint is related to the notion of a prediction interval (PI). The appropriate choice of percentile ( for instance 90 or 95% but surely not 50% ) has not yet been decided at VetCAST, but VetCAST anticipates that due to the large inter-animal variability and the paucity of data, a very wide prediction interval could have large consequences for the establishment of a CBP

Prediction interval vs Confidence interval A Prediction interval (PI) is a range that is likely to contain a specified proportion (e.g. 90%) of the population. PI is a random variable and it has its own confidence interval (e.g 95 or 99%) Same conceptual difference as between a SD and a SE

PTA mean vs quantiles Breakpoint for amoxicillin in pigs (CLSI) Q_91%: 0.125mg/L Q_59%=0.5mg/L NO PK variability (mean): 0.5mg/L

How to compute PTA From already published PK parameters The CLSI approach From raw data (meta-analysis) Recommended by VetCAST

Florfenicol in calves as an example

Florfenicol Three data set 50 calves Rich data

Computation of the PTA by Monte Carlo Simulations (MCS) PK raw data From clinics & preclinical unbalanced data Population modeling Monte Carlo Simulation (n=5000 animals) PTA(%)

PTA using MCS We need a model (exponential) to simulate the 5000 curves

Why we need a population modeling To aggregate sparse and unbalanced data Data collected in clinics From different companies/lab The only fair approach for a meta-analysis To document covariates to generate expert comments (rules) Formulation effect? Breed effects? Preruminant vs ruminant cattle ?

Florfenicol: Unbalanced data Three data set 50 calves Rich data LOQ of 0.25 (RED) vs 0.05 (Blue and grey) µg/ml

Unbalanced data for florfenicol 3 data sets & 2 models! JVPT 2011 What about T>MIC for low MIC? Other source

Unbalanced data for florfenicol 2 data sets & a single POP model

IM administration 30mg/kg

Clinical cutoff

The clinical CO (COCL) The COCL is based upon the collection of isolates obtained during the clinical effectiveness studies. COCL reflects the upper limit of the MIC values associated with a high likelihood of clinical success [probability of cure (POC)]. There is no set method for establishing the COCL, and no hard target for POC. POC=0.9 POC= Probability of cure MIC/Dose

Issues for to determine the clinical cutoff No  public data relating MIC to outcome Rem: EMA, not FDA require to sample isolates during clinical trials An assumption that needs to be consolidated It exist a relationship between MIC and clinical outcome or between AST results (S,I,R) and clinical outcomes

Conclusions and Clinical Relevance—Recovery of tilmicosin-resistant M haemolytica or P multocida isolates was rare, and no association was detected between MIC of tilmicosin and treatment response

MIC vs. clinical cure for sensitive strains (n=160) for a quinolone

AST results and clinical outcomes (Merged results for 4 AMDs from 3 different classes) Percentage AST cure failure improvement S (n=364) 72.5 19.5 8.0 I (n=23) 69.6 21.7 8.7 R (n=17) 52.9 41.2 5.9 Pulmonary conditions; Pasteurella multocida and Manheimia haemolytica

Decision step: The CBP

Decision steps VetCAST will follow EUCAST procedures involving transparency consensus (not a vote) Independence For VetCAST, as currently is the case for EUCAST, the veterinary pharmaceutical industry will be a partner and will have an active consultative role, but will not have a financial role or participate in decision-making.

Decision step There is no generally accepted formula or mechanism for combining these different pieces of information into a single CBP to define the Susceptible category. Decisions in the final analysis are a matter of judgment by experts drawn from several disciplines and based on reflective consideration of all available information. in addition any other factors, scientific or non-scientific, that are relevant to practical conditions, to ensure harmonization between EU countries will be also considered

Clinical breakpoint vs ECOFF

CBP and dosage regimen Operationally, if the PK/PD breakpoint is below the ECOFF, it probably means that the current dosage regimen for that AMD is too low to treat the wild-type population. In this case, VetCAST will not establish a CBP dividing the wild-type MIC distributions

Diameters vs. MICs Diameter is a surrogate of MIC What is the value of this surrogate? Convenience Reliability? For harmonisation

For a quinolone: MIC vs. Diameters Diameters (mm) MIC: µg/mL MIC should be the independent variable when discussing this relationship

Expert comments To assist clinical microbiologists in preparing their report, VetCAST will provide ad hoc guidance documents for interpretation of ASTs. This will ensure the adequate and contextually correct interpretation of AST results in the light of the animal’s local or regional circumstances.

Conclusion Position paper “En route towards European Clinical Breakpoint for veterinary antimicrobial susceptibility testing: the VetCAST approach”