14.2.1 Disease and Response Similarity Between Adults and Children With UC and CD

The pathophysiology of IBD is similar in adults and in children. IBD is common in children and adolescents; about a quarter of patients with ulcerative colitis and Crohn's disease present before age 20 years [3,4], with incidence of both diseases rising in children <10 years of age [5]. In pediatrics, the diagnosis of inflammatory bowel disease unclassified (formerly, termed indeterminate colitis) ranges from 5% to 30%, though this diagnostic category may be overused [ 3 6–8]. In over 80% of children with UC, disease affects the colon proximal to the splenic flexure, which is more extensive involvement than most adults at diagnosis. In children with UC, there can be relative rectal sparing and patchiness that is uncommon in adults [9,10]. Growth failure at diagnosis is common in pediatric Crohn's disease, reported in ranges from 10% to 56%. In UC, growth failure at diagnosis is less common (ranging from 0% to 10%) [11].

Very early onset inflammatory bowel disease (VEOIBD) is a label used for children diagnosed with IBD who are less than six years of age. Monogenic causes of VEOIBD have been identified in a subgroup of these children, including mutations in IL10RA/B, XIAP, and ADAM17, among others [12]. Children with VEOIBD is more likely to have mild disease at diagnosis, present with a colitis phenotype; change to an ileocolonic Crohn's disease phenotype later in childhood may occur [13]. When a known mutation is identified, there is a potential for personalized therapy, with bone marrow transplantation being curative in some cases. However, when no mutation is identified (the majority of children with VEOIBD), these children are generally treated similarly to older children with IBD.

Goals of therapies in children are similar to those in adults induction and maintenance of remission. However, there are some important differences. One is the evidence that early treatment with an anti‐tumor necrosis factor (TNF) biologic in severe inflammatory Crohn's disease leads to better clinical and growth outcomes than immunomodulator monotherapy, with the caveat that this may not be the case for children with a stricturing phenotype at diagnosis [14,15]. Another difference is the length of time a child lives with IBD, and the number of years a pediatric patient is faced with disease/treatment. Thus, avoidance of treatment failures and durability of treatment is of primary concern, and particularly the impact of immunogenicity and serum levels (exposure) on durability [16,17]. IBD and therapeutics used to treat IBD may impact growth and development in children and adolescents. Chronic inflammation in Crohn's disease causes growth failure and osteopenia in children and adolescents. In addition to the effects of chronic inflammation, glucocorticoids are known to suppress bone turnover in growing children via a combined effect of osteoclast stimulation, and chronic impaired bone formation; bone turnover is a necessary process for bone modeling and growth [18,19]. Conversely, anti‐TNF therapies have been shown to improve bone turnover, bone density, and growth in children with Crohn's disease [20]. In addition to effects on growth, there are other disease and therapeutic safety concerns that may be different in children and adolescents, such risk of lymphoproliferative disease and hemophagocytic lymphohistiocytosis due to lack of prior exposure and immunity to EBV when using thiopurines [21]. Special care must be taken to monitor for side effects that may be unique to children and to understand that pharmacokinetics (PKs), as well as the exposure–response relationship, may differ in younger children.

In the small number of randomized controlled trials (RCTs) of therapeutic proteins evaluated for regulatory approval in pediatric IBD, responses to infliximab (IFX), and adalimumab were generally similar to those in adult trials, with the exception of a lower maintenance of remission in pediatric patients with Crohn's disease treated with adalimumab (see Table 14.1). In the REACH trial, children with moderate‐to‐severe active Crohn's disease unresponsive to immunomodulators (mercaptopurine or methotrexate) received open label infliximab for induction, and then were reassessed after three doses of the medication. Children that responded to treatment were then randomized to receive a standard variable weight‐based dose of infliximab (5 mg kg⁻¹) either every 8 or every 12 weeks [24] . In a study of children with moderate‐to‐severe ulcerative colitis, children received open label induction with three doses of 5 mg kg⁻¹ of infliximab, and then were assigned to receive maintenance doses every 8 or 12 weeks [22] . In both studies, efficacy was assessed by response and remission, as measured by clinical disease activity indices, and in the UC trial, by endoscopic assessment for mucosal healing. Both studies demonstrated that induction dosing led to response in the majority of patients, and that patients randomized to every 8 weeks had superior remission rates to those randomized to every 12 weeks of therapy. It is important to take into account that children enrolled in these trials were taking (and failed) immunomodulatory therapy, and studies were done prior to the understanding of therapeutic drug monitoring. As such, response and remission rates were achieved when subjects did not have a clinical need for dose escalation [22] . Twenty‐five percent of subjects in the pediatric UC trial discontinued therapy for lack of response before Week 8 [22] . As more rapid dose escalation and therapeutic drug monitoring become increasingly utilized, these rates of response and remission are likely to be different.

In a study done to obtain regulatory approval for adalimumab, children received open‐label adalimumab in addition after failing immunomodulators, and patients who responded were randomized to continue adalimumab in a maintenance study (Table 14.2). The adalimumab dose was given as a fixed dose based on body weight. In this study, high and low doses were used for maintenance, with an overall remission rate of 33.5%. Subanalysis determined that patients who had previously been treated with another anti‐TNF agent (infliximab) were less likely to achieve remission [26]. A similar study is currently underway in ulcerative colitis; the study has had difficulty recruiting, in part because of the inclusion of a placebo arm in the maintenance phase.

14.4.1 Dose Selection

The identification of an appropriate dosing regimen remains a major challenge in pediatric product development. For TPs, investigators typically have limited PK and pharmacodynamic (PD) data in the pediatric population. For this reason, adult PK data is often leveraged in order to determine a pediatric dose. Population pharmacokinetic analyses that utilize both adult and pediatric data are routinely used to determine appropriate dose in pediatric patients. As discussed previously in this chapter, in situations where extrapolation of efficacy is applied, the pediatric dose may be selected to obtain similar exposures in adults that have been shown to be safe and effective for the same disease. Dosing approaches for TPs in children also need to take into consideration the rapid growth in maturation in pediatric patients as well as potential differences in immunogenicity as compared to adults [28]. Common approaches include dosing related to body size (i.e. body weight or body surface area [BSA]‐based dosing) and age. There is limited data on the disposition of TP particularly in young children (for e.g. less than two years). However, given that IBD trials are conducted in patients two years and older, a stepwise approach of dose selection where PK is collected in older children to be able to perform reasonable exposure predictions in younger children is not routinely necessary.

14.4.2 Exposure–Response Analysis

Exposure–response (E–R) analysis is routinely utilized to make drug development, therapeutic or regulatory decisions [29]. A well conducted E–R analysis for efficacy and safety provides evidence for rational dosing recommendations. A significant exposure–efficacy analysis also provides supportive evidence of effectiveness. While E–R is a useful tool to utilize individual level PK and efficacy data to define the therapeutic window and hence select an optimal dose and regimen, it is also important to appreciate that data used as input for E–R analysis drives the decisions one can or should take from these analyses. For example, if data from only one dose level is available, conducting exposure–response analysis has little value. If the aim is to understand the effect of various exposure‐metrics on response, data from various regimens which can differentiate the effect of exposure‐metrics (e.g. C_max or C_trough) on response has to be included in the analysis. It is also important to note that conducting informative adult programs is important for design of future development programs in children. A well‐established dose or exposure–response relationship in adult patients helps define the dose in the pediatric population. In addition, collecting PK along with efficacy data in adult and pediatric patients enables the use of exposure–response analysis to support dose selection. Moreover, in the context of extrapolation of efficacy in children from adults, similarity in exposure–response between adults and children is not a prerequisite to successful pediatric approval and labeling. In other words, in the absence of comparison of exposure–response between adults and pediatrics, a well‐designed pediatric program on its own can provide information on evidence of effectiveness and support dosing in children.

There are two TNF‐α blockers approved in children for IBD, infliximab for Crohn's disease and ulcerative colitis while adalimumab is approved for CD. Exposure–response analysis played an important role in approval of infliximab [30] for UC and adalimumab for CD [31]. These two case studies are briefly summarized below.

14.4.3 Therapeutic Drug Monitoring

Therapeutic drug monitoring (TDM) was first inserted into the IBD clinical management algorithm when thiopurines were considered the most effective maintenance therapies for both CD and UC. Taking the lead from the oncologists, GI physicians studied the break down products, also referred to as metabolites, and their associations with both efficacy and safety. The multiple studies supported dose escalation in the face of subtherapeutic drug concentrations. There is a fine balance, however, between too little and too much drug as very high metabolite concentrations can compromise the safety.

The principles of dose optimization and TDM were then applied to an anti‐TNF monoclonal antibody (mAb)‐infliximab. Although the class of therapy was new and familiar principles, it took many years for the IBD community to understand how to interpret the data on TDM with IFX. The initial reports pointed to the notion that any amount of measurable drug above zero was associated with better clinical, biochemical, and endoscopic outcomes [35]. Over time therapeutic ranges with a minimally effective trough concentration were introduced, typically 3 or 5 μg ml⁻¹ depending on the assay methodology used and collected in the maintenance phase of IFX treatment, often after many years of exposure [36]. With the exception of a few studies, most of the data comes from cross sectional studies that evaluated patients either in remission or relapse and identified a median level that is associated with the state of disease activity. This reactive based approach was predominated the literature until more recent years whereby efforts have been made to identify therapeutic drug concentrations earlier on in treatment such as in the induction phase or just postinduction and dose escalate accordingly [ 16 37–39]. This proactive dosing strategy has demonstrated not only enhanced efficacy but also perhaps more importantly reduced surgeries and hospitalizations in IBD patients. Not unlike the majority of the previously published studies, this optimization occurred in patients during the maintenance phase of treatment and in some who have had years of IFX exposure.

Proactive dose optimization would perhaps be most effective and meaningful if performed within the induction phase of IFX therapy. The reasons for this are plentiful. IFX, like other mAbs, is cleared at a faster rate in the face of a high inflammatory burden which would be more so the case at the initiation of therapy as the majority of IBD patients are started on IFX for active inflammation failing conventional therapy. In the face of rapid drug clearance, patients are at highest risk of developing anti‐drug antibodies (ADAs) which most often results in the discontinuation of therapy. Up to 20% of patients with acute severe UC already have measurable ADA six weeks into therapy when receiving standard outpatient doses that under most circumstances would be considered inadequate for such a high burden of inflammation [40]. There exists a “chicken or the egg” phenomenon as to whether lower drug concentrations condone the development of ADA or whether ADA presence results in lower drug concentrations. Short‐term mucosal healing rates were significantly improved when drug concentrations during induction were quite elevated compared to the accepted maintenance concentrations between 5 and 10 μg ml⁻¹ [41]. Thus, the ideal approach to optimizing exposure early would be to apply TDM during induction and adjust dosing accordingly. Unfortunately, there remains a paucity of data on what would be the ideal dose and frequency beyond the on label 5 mg kg⁻¹ induction dose administered at an interval of 0, 2, and 6 weeks. This guess work, however, may be able to be overcome by employing dashboard‐based PK predictions employing Bayesian PK algorithms to forecast individualized concentration time profiles to determine optimal dosing expected to maintain target trough concentrations.

Dashboard systems have the potential to help clinicians adjust dosing over the course of treatment by using modeling to personalize dosage [42]. The first study to explore dashboard predictions in IBD practice, reported that when using Week 14 clinical data only, the dashboard recommended either a dose or an interval change (greater than a 0.5 mg kg⁻¹ or more than one week difference) in 43/50 patients; only 44% recommended to have standard of care (SOC) administered dosing (combination of 5–10 mg kg⁻¹ and an interval every six to eight weeks). When Week 14 IFX concentration and ADA status was added to clinical data, dose and/or interval changes was recommended in 48/50 patients, and SOC dosing was recommended in only 22% [43]. The possibility that target concentrations will change with a more defined treat to target approach in IBD patients must be considered. The IFX PK dashboard can adjust dosing depending on the selected concentration target. Clinicians are limited to on label dosing in general, and third‐party payers have evolved to more flexible dosing approvals. That being said, outside of a patient with rapid clearance, all outputs are within label or generally accepted by the payers. With a PK dashboard, GI clinicians can improve upon current dosing choices and minimize ADA formation for the IFX treated IBD patients. The use of these systems will need to be verified by prospective clinical trials showing the benefit of this approach, and physician and patient education utilizing these systems will have to be made available.

14.4.4 Adaptive Designs

Since the FDA issued the draft guidance for industry on adaptive design clinical trials for drugs and biologics in 2010 [44], an increasing number of trial sponsors have conducted clinical trials using some type of adaptive designs to help accelerate the drug development process. Adaptive designs typically involve four types of adaptation, including sample size re‐estimation, seamless Phase 2/3 designs, group sequential designs, and enrichment designs [45]. Sample size re‐estimation is the most widely used adaptive design in both blinded and unblinded trials. Although the adjustment of the Type I error rate may not be required for the blinded sample size re‐estimation, the benefit of sample size increases can be limited in this situation. When sample size re‐estimation is to be conducted based on unblinded interim data, analytical approaches to controlling for Type I error need to be planned prospectively. Although sample size increases can mitigate the lack of study power, they do not guarantee successful trials. One failed case was observed for the assessment of memantine in children with autism.⁵

Another commonly used adaptive design is the seamless Phase 2/3 design. By directly rolling over patients from Phase 2 to Phase 3 and combining these two‐phase trials into one larger trial, not only can the sample size requirement be reduced but also the lead time between the two phases can be curtailed, thereby shortening the entire product development duration. The Phase 2 portion of the entire study can serve different purposes such as pharmacodynamics evaluation besides efficacy evaluation. This type of Phase 2/3 seamless design is an appealing option for pediatric trials that face the challenge of patient enrollment. One well‐known example is the successful trial of propranolol for infantile hemangiomas [46]. As adaptive designs have been gradually utilized in the pediatric trials for trial efficiency, it is necessary to ensure trial validity and integrity by developing a prespecified statistical analysis plan to control Type I error and forming a data monitoring committee to handle potential interim data unbinding before trial initiation.

14.4.5 Advantages and Disadvantages of Using External/Historical Controls

Performing controlled studies in pediatric IBD is at times heralded by multiple practical and ethical challenges. Placebo may be associated with major ethical and feasibility barriers while choosing an active comparator usually requires a sample size too large for a pediatric study. Therefore, benchmarking results of uncontrolled pediatric study with other sources may occasionally be considered as a valid option. One of the possible study designs include comparing pediatric outcomes of open label intervention studies with those achieved in the placebo arm among prior adult study, as recently done in the Phase 2 study of golimumab in children with UC [47]. To make such a comparison reasonable, the study design and outcome measures used in the pediatric study should be as similar as possible to those used in the prior study. Alternatively, using historical controls allows a comparison of the same age groups especially when clinical practice has been changed and concurrent comparison may be unethical or not feasible. However, the limitation of using a historical comparison including factors that may be altered over time in the management protocols, including unknown and thus unmeasured confounding variables will need to be considered. Therefore, extra care should be provided on identifying possible confounding variables different between the two time periods other than the intervention under study. A second concurrent comparison group may also be considered to benchmark the difference from the historical controls.

14.4.6 Real‐World Evidence

RCTs are considered the most appropriate study design to elucidate treatment effect of any intervention. A well‐performed and concealed randomization process is the ideal protection against hidden biases and unknown confounding variables. However, traditional RCTs may not represent “real‐world” conditions and lack generalizability due to stringent selection criteria [48]. In a retrospective cohort study of adult patients with moderate–severe IBD, Ha et al. reported that only 31% of 206 IBD patients would have been eligible to participate in any of selected published RCTs of biologics [48] . Patients would have been excluded because they had stricturing or penetrating CD, took high doses of steroids, had comorbidities or prior exposure to biologics, or received topical therapies. After applying the eligibility criteria on the entire local population, the patients' outcomes differed between those would have included in the trial and those who are typically excluded, most notably the sickest patients. Publishing real‐life outcomes could reduce the magnitude of the selection bias in RCTs but still most real‐life cohorts come from tertiary academic centers, with inherent referral bias. Nonetheless, the strength of real‐life cohort studies is the ability to collect explicit data on all treated patients within a given cohort.

Population‐based cohorts of IBD patients have a tremendous potential to facilitate our understanding of the disease, while overcoming selection bias. However, population‐based cohorts are not often available on a patient‐level and thus administrative databases are commonly used. For performing administrative research, it is first fundamental to develop and validate rigorous algorithms for case ascertainment in order to accurately identify true IBD patients within the databases [49,50]. Such algorithms may include a combination of health administrative codes (e.g. International Classification of Disease [ICD] version 9), medications and procedures. In the epi‐IIRN Israeli administrative IBD database, one‐third of those labeled as having IBD by any health contact did not have the disease as found in the validation process. Administrative research is limited by the fact that it includes only data recorded as part of clinical practice and individual patient data are often lacking. Efforts to establish disease registries and initiatives to address these challenges are ongoing (e.g. Improve Care Now, Children's Registry for the Advancement of Therapeutics).

Given the pros and cons of each study design, a combination of data arising from RCTs, patient‐level cohort studies and administrative research will likely be needed to provide a complete picture of the therapeutic effect of a given intervention. One of the commitments as part of PDUFA VI is enhancing the use of real‐world evidence in regulatory decision‐making and particularly how it can be used to support effectiveness of medications [51]. This is a great opportunity that one can leverage for accelerating product development in pediatric IBD.

14.4.7 Quantitative Systems Pharmacology

Significant advances in data measurement technologies and capabilities, and advancements in computing, are gradually transforming the biological and clinical sciences into data and computational sciences. These opportunities in leveraging computational and data sciences can be harnessed, as biological and clinical knowledge on disease and drug perturbations to a healthy homeostatic biome can be translated into mathematical equations and computer models. With the appropriate investment in the computational sciences, these computer models can be subsequently used for in‐silico drug discovery (e.g. identification of novel targets) or drug development (e.g. virtual human trials). We briefly highlight two advanced modeling approaches, namely quantitative systems pharmacology and clinical trial simulation, and describe how they can be leveraged in advancing important novel therapeutics for the benefit of patients with inflammatory diseases like IBD, where population size is small.

Quantitative systems pharmacology (QSP) is a mechanistic modeling tool that links molecular mechanisms of disease and drug to biomarkers and clinical endpoints used for assessment of disease and therapeutic effect. QSP is suited to understanding the system‐level response to treatment across multiple PD markers and clinical endpoints, and to assessing patient variability on a mechanistic basis. There has been many QSP models published that address discovery and development questions across a variety of therapeutic areas, e.g. cardiovascular, cancer, immunology, oncology, and rare diseases [52–60]. Some example applications include evaluation of mechanism of action of a new molecular entity, prediction of response in new populations or novel dosing paradigms, advancing hypotheses for mechanistic basis of response/nonresponse, and exploration of mechanistic basis for synergies in combination therapy approaches.

14.4.8 Clinical Trial Simulation

Clinical trial simulation (CTS) has been an invaluable computational tool that leverages an underlying PK–PD model or prior knowledge on exposure–response relationship, and uncertainty in that relationship, to simulate the probability of success of various trial designs, typically for a late phase design strategy. There have been many simulation studies published, describing a wide variety of CTS applications in drug development [61–65]. Examples include simulating scenarios of different potential exposure–response relationships and impact of probability of success of different designs, power evaluation for designs comparing competing drugs in the context of sparse data, or evaluating the impact of variability or uncertainty associated with PK–PD model parameters on competing trial designs for a Phase 2b study, and multi‐study or program trial simulations, bridging various assumptions on exposure–response, based on prior knowledge from literature or clinical studies.

Product development of complex inflammatory diseases that involve interaction of the environment and genetics as well as small populations is challenging. Identifying biomarkers for achieving POC, and appropriate doses that could address multiple underlying phenotypes for late stage development is complex. QSP modeling provides a mechanistic framework for integrating data on mechanism of the drug or biological product and disease and linking to relevant endpoints for the disease. The models can, therefore, be useful in proposing biomarkers of response, or stratification biomarkers, by representing a spectrum of disease phenotypes. Given a small population size, and potentially wide variability across a wide spectrum of disease, CTS can be used to propose optimal design of POC and late phase studies, based on variability in exposure–response relationship and associated covariates that may be taken into account for the design. CTS can also be useful in leveraging historical control data based on published data, or available registry data, where patients may be matched by disease relevant criteria.

In the case of pediatric extrapolation, QSP modeling has been used in supporting the case for similarity of disease, by evaluating the impact of potential differences in disease pathophysiology on relevant biomarkers of severity [60] . QSP modeling has also been useful in quantitatively evaluating the response similarity between adult and pediatric populations. Pediatric and adult virtual patients generated based on clinical trial, and registry or disease progression data, are used in evaluating the extent of overlap in key biomarker and clinical endpoint response to therapy in these populations. These simulations can take into account the progression and severity of the patient, as well as pathophysiological characteristics of the disease, and known potential differences between the two populations, thus enabling a thorough and quantitative response–similarity evaluation [60] .

Given a small population size, and potentially wide variability across a wide spectrum of disease, CTS can be used to propose optimal design of POC and late phase studies, based on variability in exposure–response relationship and associated covariates that may be taken into account for the design. CTS and Bayesian approaches have been useful in leveraging historical control data based on published data, or available registry data, where patients may be matched by disease relevant criteria.