welcome to module 4 the foundation of modeled informed drug development M and modeled interconnectedness this presentation will provide you the foundation knowledge of MD to other pieces of training in the series of model interconnectedness the M model interconnectedness series are developed by adim Taylor where Josh Krishna and Amy Chong who are senior directors at satara and in collaboration with critical path Institute in the next few slides we will talk about the foundational principles underlying model informed drug development or MIT there are many drivers of uncertainty within drug development these include confidence in targets confidence in the drug in question confidence in The Chosen end points and confidence in regulatory decisions so mid can be viewed as a rational approach to effectively and efficiently accelerate drug development and as you can see in this inset these are the five different objectives that you um see and that can range from various components of these drivers including pathway way Target drug risk benefit and payer perspectives and within each pathway there are tools and techniques that can help answer the question whether this is the right Pathway to pursue right target to consider right molecule to develop are we picking the right dose and do we have enrolled the right patients um when looking at this in an interconnected manner they convey a story that underlines the drug product development ecosystem in this slide we reflect on how model informed drug development can help with key development decisions as you look towards the Continuum of Discovery preclinical early clinical late clinical and Commercial which is the horizontal arrow that you see on the top of the slide there are very specific questions that one needs to take into account within each phase of development for example within Discovery the key questions we need to answer is do we know the target do we know the metabolism transport of the drug in question can you design the molecule outside of the patent landscape that can differentiate you from a competitor for example what is an Optimum synthetic pathway what's the risk of of Target effect so these are some of the questions within Discovery phase As you move towards preclinical questions are more tailored towards do we have the right information to consider before going to early clinical for example what compound to pick what should be the first in human dose are there likely to be drug interactions what's the risk of cardiac safety and do we have a biomarker that can best Define the exposure response strategy with an early clinical the questions are separate and perhaps interconnected with both preclinical as well as late clinical this includes do we have the right trial designed to show proof of concept are we testing the best dose what is the risk benefit profile and are there subpopulation of interest what is the efficacy and safety of combination therapy that could also be pursued within early clinical phase and this is also a time when you have preliminary idea of how good of a product you have relative to either a standard of care or a competitor product and finally a very critical question is have we made the right investment decision because once you move into late clinical phase and you start work within that phase the risk of making a wrong decision could be a financial challenge some of the questions in late clinical for example include should we continue with this development program what is the therapeutic window for this drug how can we optimize the study design what is the optimum inclusion exclusion criteria which could inform labeling how do we price a drug do we have the right M Market access plan what is the strategy for Pediatric development as you get into commercial space these questions have more proximal benefit for approval but also launch and these include benefit risk assessment real world evidence about return of investment uh payer access perspective what additional indications to pursue should we consider life cycle management and so on so forth again within these questions you already can see varying degrees of interconnectedness of these models to allow M to influence various decisions during the different phases of R&D it is essential to implement an M strategic plan the development of the M strategic plan is supported by the five different M tools the first step in the plan development is to identify the key R&D questions where MD will provide impact there's seven different themes of questions that can be asked concerning the compound mechanism and disase or indication levels to expand the three key levels are disease level questions and activities that can be answered in advance of the development of any particular compound compound level questions and activities that focus on the data associated with a compound of Interest additional mechanism level considerations compound leveled or disease level questions and activities where there is a focus on the mechanism of action and a knowledge gained from other compounds with a similar mechanism of action for the seven themes they are the medical need or commercial viability R&D question related to the understanding of the medical need and the areas of potential differentiation from the standard of care for particular diseas or indication this can inform the likelihood of a particular compound achieving the important aspect of a product profile at each stage of development efficacy R&D questions related to the characterization of the dose exposure response relationship for important efficacy outcomes safety or tolerability are indeed questions related to the characterization of the dose exposure response relationship for important safety or tolerability outcomes pharmacokinetics R&D questions related to the characterization and extrapolation of the pharmacokinetic properties of a drug across species and patient populations the general expected impact of a progressive disease State intrinsic for example age organ impairment or extrinsic factors for example co-administrated drugs and the influence of formulation or Administration method on drug exposure benefit or risk are the questions related to the definition and quantification of the relative tradeoffs between important efficacy and safety outcomes to determine optimal dose regiments they're sufficiently effective and safe clinical viability R&D questions related to the assessment of potential development programs for a particular indication considering options concerning populations subpopulations inclusion or exclusion criteria study design R&D questions related to the optimal design of the subsequent studies balancing the cost and the time of the current study versus anticipated future risk given the predicted confidence in achieving the required product profile once the key questions identified the proppr modeling approach or combined approaches which is our model interconnectedness will be selected along with the Assumption settings and EV validation strategy and finally the impact level assessment will allow us to evaluate how the mid impacted the decision making from both industry and Regulatory perspectives the impacts would petitioned into three categories descriptive Justified which means decision informative and evidence substitution different key questions can be asked at different stages of drug Discovery and development process from Target selection and validation non-clinical phase phase clinical phase to life cycle management if we choose one of the themes as efficacy you can see different questions can be asked for disease level compound level and additional mechanism level different types of quantitative approaches can be applied to answer them which allow placing the puzzle together for example at the disease level the question would be what are the dose response relationship for the compound used to treat the disease a comparator database can be built from the literature for known compounds or perceivable used in the disease or indication of interest to support Explorations and modelbased meta analysis and BMA previously we mentioned assumption setting and validation strategy is part of the fundamental key step for M especially when the data and information are spars there are five types of important assumptions namely pharmacological assumptions physiological assumptions disease assumptions data assumptions and mathematical and statistical assumptions the key is the Assumption will need to be clearly identified and documented prior to any M processes as the Assumption will be tested and evaluated if testable which will impact the next modeling steps and decision the assumption is also very important for all types of modeling for example in empirical modeling as the model is quite simple many assumptions need to be made for example the number of compartments based on the profile for mechanistic model as the model is quite complex assumptions needs to make for example prior knowledge of certain biological pathway processes rates from the literature for extrapolation approach the Assumption must be both presented in the plan and also in the report of how the assumptions can be validated and test sensitivity analysis is also a very important way to validate assumptions again assumptions should be clearly stated and documented at the submission stage or when interacting with internal stakeholders external investigators or R Regulators patient groups because this will allow more transparencies to the uncertainty and certainty and allow better discussions and communication the M impact level allows evaluation of how the quantitative of approaches impact decisions which varies from the audience and the approaches used the classification range from low medium to high it is a system to rank how impactful a piece of quantitative work is to internal decision making load impact is when the work is used to describe the data so moderate in terms of decision fitting data to a PK model at the end of the study analysis medium is when the work can influence a decision for example population selection study design Etc using simulation or clinical trial simulation hide is based on when not additional data or experiment is required to impact on decision for example using extrapolation to wave an unnecessary clinical trial this system also ensure we can classify the type of work and the associated power of MIT with the modal interconnectedness it allows knowledge to keep building up and knowledge keep propagating and this will increase the probability of hide impact work to decision next we will discuss the types of models that can be considered during drug development there are different ways to classify models the classification system I will use is based on the white paper on good practices in modeling by the European Federation of pharmacutical Industries and associations model informed drug Discovery and Development workg Group the first type of model we'll discuss is the empirical dose exposure time model this type of model relates doses or exposures such as the area under the concentration time curve or other summary measure of drug exposure to the pharmacodynamic response and there could also be a component of time this type of model uses simple empirical functions to describe the drug effect for example it could be a linear drug effect it could also be a hyperbolic drug effect model more commonly known as an emac model and in this type of model there is a limit Beyond which the drug effect can no longer increase even if drug concent ations increase as and another model is the exponential type of model this is not an exhaustive list but just an example of the types of simple empirical functions that are used which are not mechanism based this type of model uses limited assumptions and as a result it has a limited extrapolation potential the example on the left is a model for Alzheimer's disease that describes the change over time time in the Adas cognitive score which is an efficacy Endo for Alzheimer's disease at different doses of the investigational drug and so it looks as the change over time so this would be a dose time model in the second example the drug response is related to the drug exposure in this case the area under the curve using a saturable or emx model and this can be used to predict the response rate range for different dose levels as you can see is being done on the curve and So based on this type of model you can read of the range of exposures at the different dose levels which are represented by the horizontal bars near the bottom of the graph and then you can read of the response rate um for each of those uh do or exposure ranges the next type of model is the semi- mechanistic pharmacokinetic pharmacodynamic or pkpd model this type of model is based on what we know about the biological and pharmacological mechanisms of the drug and can use data from different sources it can use human data that is clinical data or preclinical data which would include animal data as well as inv Vitor experiments and these can be used to add mechanistic interpretations to the model parameters we can think of the types of parameters of the models as consisting of the drug parameters and the system parameters the drug parameters would be those parameters that are specific to the compound for example the pharmac the pharmacal kinetic parameters as well as the drug effect parameters such as the drug potency the system parameters would describe the system that is being modeled for example the a synthesis rate of a biomarker being modeled these models are more complex than your empirical pkpd models and also have somewhat of a mechanistic basis and as such they can be used to extrapolate beyond the data that were used to model the um to create the model the example I'm showing here is of a population pkpd model for serum intact parathyroid hormone concentrations or ipth which would be our biomarker end point this model shows one of the simplest types of semi mechanistic pkpd models which is called the turnover model it shows how the plasma concentrations of the drug impact the synthesis rate of ipth and so here you can see that there's a synthesis rate of ipth and also an elimination rate of ipth which under uh circumstances where there's no drug intervention balance out to maintain homeostasis but when you add the drug concentrations the drug then reduce the reduces the synthesis rate of ipth here we get a glimpse into the interconnectedness of models as we see how the pharmacokinetic model is used to predict the concentrations that will drive the pharmacodynamic effect in the graphs below we see the mean model predictions shown as lines overlaid on the mean observations which are shown as the Open Circles the left graph shows the drug concentration profiles at various doses which are generated by the PK model and on the right we see the model predictions of the hormone concentrations over time at the different Doses and as we can see here this model does a good job of predicting the concentration time Prof profiles of ipth next we come to modelbased metaanalysis this type of modeling approach uses summary statistics such as mean median standard deviation of trial data to assess the comparative risk benefit ratios of different compounds of Interest previously we have seen the other models looking at individual level data these models look at trial level data and they typically use data found in the literature or in public databases these models will account for the impact of the treatment but not only that they will also account for differences in the patient populations across the different trials as well as trial characteristics they can use empirical models or they can use models that account for the clinical pharmacology of the drug and they can also be used for extrapolation if they're part of a disease progression model and another application is you being used in estimating scaling factors this example shows how a model-based metaanalysis was used to describe and predict the response over time to various osteoporosis drugs which are shown in the legend on the right and their effect on increasing bone mineral density on the left you see the effect of the various drugs in increasing Lumar spine bone mineral density and on the right total hip for mineral density and these drugs these models can be used to POs to see how to position the drugs under development and to U compare them to drugs that are already on the market now we'll talk about quantitative systems pharmacology or qsp models and physiologically based PK or pbpk models these models are being grouped together because they are similar in that they are both bottom up models what this means is that these drugs use an understanding of biology physiology pharmacokinetics pharmacological processes and the pathology of the drug to predict drug exposure as well as the drug effects which could be either therapeutic or toxic effects so these drugs are based on what we know about the systems consideration unlike the other models which we have looked at which are typically looking uh using the data to develop the model structure and so in those in that sense they are considered to be top uh top- down models pbpk models and qsp models need data from a variety of sources for example they will have uh data on blood flow rates which is based on the biology or physiology of the system and they could also include data about the drug binding characteristics and also the drug physical chemical properties such as permeability or solubility these are multikill models so they Encompass the target level up to the cellular level and ultimately up to the whole body level and as a result of their complexity they consist of systems of many differential equations the example I'm presenting here shows a pppk model for monoclonal antibodies as you can see the model describes the monoclonal antibody transport and partitioning into the different organs and plasma which are represented as compart ments the model relies on blood flow rate and blood flow rates in the various organs as well as reaction and binding at the tissue level as you can see here because there are many compartments this model will consist of many more ordinary differential equations than your typical empirical pkpd model or semi- mechanistic model due to their bottom up nature these models are more often used for predictions rather than to fit data although predictions are typically compared to the observations in order to validate the model as we can see from the plots on the right the pppk model was able to provide good predictions of concentrations in the plasma and in different organs of knockout and wild type mice epidemiology models describe the spread of disease and the duration of epidemics and they show how individuals in a population move from one status to another for example they could move from a a status where they are susceptible to an infection to infected and to a recovered status they can use either ordinary differential equations or be based on a stochastic framework and they also are not limited to experimental studies but they can also use observational studies and these models can also be used from extrapolation for extrapolation and they often are the model in this example describes how individuals in the population move through various States from susceptible to influenza infection to exposed infected and eventually recovered under scenarios of treatment with osel tamere on the right we see a model fit of the fraction infected over time represents by the dash line and the obser ered fraction of population infected which is shown with the solid line and we can see that the model predictions correspond well with the data Health economics and outcome research or H models are models that provide economic analysis of Health Care interventions these types of models establish the efficacy of an intervention and compare it to the effectiveness of other interventions especially existing interventions for the indication in question these models do not only consider efficacy but also in consider the incremental cost efficiency and this is used to determine the optimal clinical application for the intervention and also its overall economic value decision makers use these types of models to help to select the best intervention for a given patient population and they are also used by Regulatory Agencies and governments to determine reimbursement structures for different drugs our example shows Health economics and outcomes research modeling to compare the cost effectiveness of pembrolizumab to the standard of care in treating patients with non-s small cell lung cancer a partition survival model was developed and this model had three mutually exclusive health States as shown in the top left figure the patient starts in a progression-free state then experiences Progressive disease and eventually death alternatively death can also be due to other causes outcomes and costs of Adverse Events were also incorporated into the model and for patients receiving each intervention the cumulative total costs and health outcomes over the time Horizon were estimated using the time in each of the modeled Health States shown at the bottom left are the clinical parameters of the model as well as a list of resource utilization and cost inputs in the next part we'll go into more details about the interconnectedness of models in the field of engineering biology and Medicine we can consider all the processes we want want to investigate who has an input a main system and an output connected this is represented by the yellow pink and green bubbles in here as of all the processes there will also be a float and propagation of information and feedback learning through multiple iterations it enabled us to establish a learning and confirming cycle if then we consider from an mid point of view view the input includes the key questions data availability Assumption of data and model disease or compound based on prior knowledge this will then fit into the types of models to be chosen and how are the different modeling approaches should be connected to develop a framework to answer the question which resulted in our interconnectedness the output can result from each step off or each type of modeling work as a result of the model interconnectedness to produce an overall impact on the decision making here I want to emphasize that the impact can happen in various stages of drug Discovery and development process ranging from the design of a clinical study assessment of a compound potential go noo decision and Regulatory submission again the interconnectedness allowing us to learn as we go along which is like putting puzzle pieces together M can apply to all stages of drug Discovery and development it increases our knowledge and propagates the information to allow us to have the full package and confidence of drug and disease the key questions and data Ste the choice of modeling approach and how interconnectedness can establish through the learning and confirming cycle assumption s will get evaluate confirm and update as the process continue which decrease our uncertainty and increases the confidence of the drug and disease a very common type of model interconnectedness is the use of population PK models to derive concentrations or exposures that are used as inputs to other models another type of model interconnectedness is multiscale models that can include different types of models in this example we see how several different types of models were used to describe the local delivery of the chemotherapeutic drug Dr rubison to tumors the top set of figures represents the physiologically based drug delivery model consisting of compartmental PK models to describe the kinetics of Dr rubison release from the liposomes the PK model is connected to the tumor compartment shown in the bottom left figure in blue figure B on the bottom right shows the detail of the tumor model on a cellular level which can be considered a physiologically based qsp model that describes the transport of doxorubicin in and out of the tumor cells and includes transport through the cell membrane and the cytool as well as reversible binding to DNA in the nucleus different types of models can be leveraged during different stages of development as shown here in this example which will be elaborated on in one of the case studies in summary the development of pisum map used different types of models as shown here during different stages of drug developments for internal decision making such as study designs and also to support registration and labeling in addition to model interconnected within a compound the interconnectedness can also apply to understand any disease here illustrate how using various modeling approaches from disease progression model pkpd to qsp to learn about the disease the mode of action of the compound how compounds of different classes are different when targeting Alzheimer's disease the safety of a compound in terms of Cy prolongation and finally how you using qsp model can provide scheduling decision during covid-19 interruptions often the output from one type of model will become the input for another model types for example with disease progression models from an individual levels data it can be combined at an analyze data from the literature using aggregator levels of data to form a modelbased meta analysis per is public applications you can see from the diagram at the bottom left corner defines parallel drug disease modeling at different stage of R&D as you can see in the x-axis and answering different key questions at different levels which is at the Y AIS such as right target right tissues right safety right patience and right commercial potential however we interconnectiveness here we propose that we can combine this to diagram and the different approaches can answer all the questions at different stages on R&D an important not is that the application of pppk and mbma don't restraint to answer certain question at a specific stage of development for example either as early stage or late stage of development instead both approaches can be used for translation and analyzing clinical data for example the usage of mbma can utilize nonclinical and clinical literature data to answer questions on tissue safety patient population and right commercial for example by having an early head-to-head comparison this slide reflects on the general theme of model interconnectedness and as we can see here there are multiple aspects to this model interconnectedness which can range from as simple as PK or pkpd or pbpk models but it can also be more mechanistically complex and that could have an effect based on the disease Focus uh wherein there are multiple disease specific platforms and quantitative systems pharmacology models that come into play as you go more into the market facing studies more questions emerge about modelbased metaanalysis clinical utility index as well as pharmacology to pair models where you have stakeholder assessment including the payer sponsor regulator and more importantly the patient so this schema reflects the modeling ecosystem which comprises of intricate models interconnected by various opportunities as the molecule transcends through various Discovery and development interfaces into the marketplace next we'll discuss the different considerations When selecting a modeling approach the level of complexity and interconnectedness will be depending on the purpose and the key questions of the analysis and the availability of data and prior knowledge for example in a clinic where we have a combination of Rich PK sampling in Phase One and Spar sampling in phase two the population PK modeling approach using an empiric compartmental modeling approach would be ideal however in a situation where physiological information is relevant in the modeling for example including the organ size differences and maturations in pediatric population we would consider the application of physiological based pbpk modeling it is important to note that the top down and bottom up approaches can be applied together instead of carrying out independently had therefore knowledge exchange and pass through the development stage of the different models an example will be using nonclinical translational pppk modeling for first time in man projections while then using the pppk model to simulate the populations in phase one after that using this simulated phase one data to develop Ved an initial pop PK modeled for optimal design for design of the PK sampling for the new trial before we embark on selecting the type of model we want to use we need to understand the goal of the moding exercise and based on that goal we can then select a fit forp purpose model that can accomplish that goal and that fit for purpose model could be something as simple as a back of the envelope calculation that calculates the dose needed to achieve a certain exposure based on non-compartmental analysis data or it could be something as complex as a qsp or a pbpk models so some of the things we need to consider are what questions do you need to answer and how will you apply the model what data do you have and what do you know the data you have and what you know about the system or the drug are going to restrict the amount the types of models that you may be able to use at the moment another important thing is how long do you have to develop and apply the models if you don't have a very long time you may have to settle for a more simplified model and importantly what resources do you have for example what kind of computational resources do you have enough computational power to develop a very complicated model also Human Resources what kind of skills do the people people who are going to be developing the model what kind of skills do they have let's talk some more about available time and resources so if you need answers quickly some of the models that you could use would be the dose exposure time models empirical PK and pkpd models modelbased metaanalysis assuming you already have a database available for you to use um as data for modeling creating your models and interestingly pbpk models the reason I put pbpk models on this list even though they're more complex is that they're typically relatively fast with dedicated software um and there's commercial software available specifically for pbpk modeling however we have to consider that pbpk models require lot of data from different sources so if additional experiments are needed to provide input into the pppk model the time that that's going to take needs to be taken into consideration if you have more time available then and it's appropriate you want to look at more semi mechanistic pkpd models or pbpk which appears again and qsp modeling a note is that semi mechanistic pkpd and qsp models can be very time and computationally intensive they're also more complex and more difficult to implement so you need to keep that in mind next we'll discuss the different types of data used in modeling a lot of the models we've discussed use individual level data or subject level data where we're taking data from particular individual subjects or animals for example if it's a preclinical data but we can also use agregate level data in certain types of models aggregate level data are data that are summarized across individuals for example this would be data summarized across individuals in a treatment arm for example for a diabetes study it could be data showing the mean change in hba1c levels over time in different treatment arms so you'll have the mean median standard deviation etc for each treatment arm these are often derived from published data but of course the sponsor can aggregate the data from their own internal studies aggregate level data are most commonly associated with modelbased metaanalysis however they're also used in health economics research and epidemiology models most of the data that we've been discussing so far are data collected from clinical trials or from planned experimental studies in animals or in vitri studies however we also have real world world data which are data collected outside of clinical trials and they can come from various sources some of them are electronic health records they could be Health Sur surveys they could even be health related apps on your mobile device or social media these types of models are used sometimes in epidemiology and health economics and outcomes research models and here I have presenting a paper by Jeffrey Barrett and Penny Heaton in case you're more interested in seeing the potential of using real world data for Global Health here I'm going to categorize data as system property data and Drug Property Data the system Property Data are the data relating to the system that's under consideration which depends on the disease of course and so those will be data that we get from our knowledge of physiology and also of the pathology of the drug and these would be things like organ size the blood flow rates to the different organs hematocrits biomarker levels and also the pathway that the drug is acting on as well as the drug Target drug properties are properties that are intrinsic to the drug itself and are independent of the system and they determine how the drug interacts with the system and these are typically gotten from invitra experiments they things like molecular weight lipophilicity permeability binding kinetics and even the enzymes that metabolize the drug as well as the fraction metabolize and the fraction eliminated by other means such as these are the inputs that usually that go into pbpk and qsp models however some some of these inputs and some of this date system and Drug Property Data can go into some more complex semi- mechanistic models extrapolation is a powerful application of mid and it is important to have a through strategy to minimize the potential unnecessary clinical studies in certain p Patric groups with existing data basic mid framework of key questions assumption modeling approaches and documentation not only allow to support extrapolation strategy but with a successful application it will support the development of regulatory guidance for example extrapolation using population PK modeling approach in partial onset CES however in the future we should increase to consider the use of various modeling approaches or interconnected approaches to support the assessment of similarity of disease and also extrapolation strategy and validation of the extrapolation strategy in this slide we bring in the various components of what we know about models as it relates to translating a clinical trial patient to a real world patient we facilitate that understanding using models one of the key components of the modeling process is translational aspects of that modeling process now within the discipline of translational Sciences you can relate an observation from a set of invitro assays progressing in into a preclinical species further progressing into a clinical trial patient now it's the challenge that we have in modern drug Discovery and development is how do you transcend finding from clinical trials into real world patients and that's where we have the discipline of Health economics and value assessment come into play for this reason it is crucial for our models that are in the early stages of development to cross talk with the models that are in play Within Market axis this is a fundamental premise of pharmacology to pair models our ultimate goal within drug Discovery and development is to understand the real world effectiveness of our therapies um one component of which comes from clinical trial efficacy and this is why there needs to be a robust translation between the patient in a clinical trial to a patient in the real world which is the underlying premise uh once again for pharmacology to payer model let's reflect on some conclusions that we have learned in this mod there are three vital observations the use of M approaches are now so streamlined that it's no longer considered novel it is an expected component of any drug Discovery and development program and when one reflects on the fact that there are more than 15 FDA guidance documents that include some form of modeling assimilation or Ms best practice it is incumbent to realize that m is a very integral component of drug development second a drug development program that does not include optimal considerations of modeling and simulation or mid is suboptimal in other words it could be a missed opportunity it's critical to understand the likelihood that the compound selected and Visa doses selected are safe and effective it is important to understand the dynamic range of the molecule with which it engages with the Target and the downstream pharmacology this not only helps identify the optimal dose range but also eliminates the chances of using suboptimal doses it's also important to understand the number of subjects required for trials are there unnecessary trials that could be reduced or eliminated M helps answer all of those questions and finally mid is a proven component of the solution where information is optimized and is financially sustainable by removing uncertainty within drug development and Regulatory Pathways it continues to evolve and its impact is only likely to grow we would like to knowledge the support and insights into these slides from the following individuals Dr Craig rer of satara Jeff Barrett of the critical path Institute and Mark selage also of the critical path Institute this presentation was brought to you by critical path Institute and satara at satara we accelerate medicines together