Artificial Intelligence for Quantitative Modeling in Drug Discovery and Development: An Innovation and Quality Consortium Perspective on Use Cases and Best Practices.

Recent breakthroughs in artificial intelligence (AI) and machine learning (ML) have ushered in a new era of possibilities across various scientific domains. One area where these advancements hold significant promise is model-informed drug discovery and development (MID3). To foster a wider adoption and acceptance of these advanced algorithms, the Innovation and Quality (IQ) Consortium initiated the AI/ML working group in 2021 with the aim of promoting their acceptance among the broader scientific community as well as by regulatory agencies. By drawing insights from workshops organized by the working group and attended by key stakeholders across the biopharma industry, academia, and regulatory agencies, this white paper provides a perspective from the IQ Consortium. The range of applications covered in this white paper encompass the following thematic topics: (i) AI/ML-enabled Analytics for Pharmacometrics and Quantitative Systems Pharmacology (QSP) Workflows; (ii) Explainable Artificial Intelligence and its Applications in Disease Progression Modeling; (iii) Natural Language Processing (NLP) in Quantitative Pharmacology Modeling; and (iv) AI/ML Utilization in Drug Discovery. Additionally, the paper offers a set of best practices to ensure an effective and responsible use of AI, including considering the context of use, explainability and generalizability of models, and having human-in-the-loop. We believe that embracing the transformative power of AI in quantitative modeling while adopting a set of good practices can unlock new opportunities for innovation, increase efficiency, and ultimately bring benefits to patients.

Assessing switchability for biosimilar products: modelling approaches applied to children's growth.

The present paper describes two statistical modelling approaches that have been developed to demonstrate switchability from the original recombinant human growth hormone (rhGH) formulation (Genotropin(®) ) to a biosimilar product (Omnitrope(®) ) in children suffering from growth hormone deficiency. Demonstrating switchability between rhGH products is challenging because the process of growth varies with the age of the child and across children. The first modelling approach aims at predicting individual height measured at several time-points after switching to the biosimilar. The second modelling approach provides an estimate of the deviation from the overall growth rate after switching to the biosimilar, which can be regarded as an estimate of switchability. The results after applying these approaches to data from a randomized clinical trial are presented. The accuracy and precision of the predictions made using the first approach and the small deviation from switchability estimated with the second approach provide sufficient evidence to conclude that switching from Genotropin(®) to Omnitrope(®) has a very small effect on growth, which is neither statistically significant nor clinically relevant.

Pharmacokinetic interactions among imatinib, bosentan and sildenafil, and their clinical implications in severe pulmonary arterial hypertension.

AIMS: This study characterized the population pharmacokinetics (PK) of imatinib in patients with severe pulmonary arterial hypertension (PAH), investigated drug-drug interactions (DDI) among imatinib, sildenafil and bosentan, and evaluated their clinical implications. METHODS: Plasma concentrations of imatinib, bosentan and sildenafil were collected in a phase III study and were used to characterize the PK of imatinib in this population. DDIs among the three drugs were quantified using a linear mixed model and log-transformed drug concentrations. RESULTS: The population mean estimates of apparent clearance (CL/F) and volume (V/F) were 10.8 l h(-1) (95% CI 9.2, 12.4 l h(-1) ) and 267 l (95% CI 208, 326 l), respectively. It was estimated that sildenafil concentrations increased, on average, by 64% (95% CI 32%, 103%) and bosentan concentrations by 51% (95% CI 12%, 104%), in the presence of imatinib. Despite increased concentrations of co-medications, treatment differences between imatinib and placebo for change in 6 min walk distance and pulmonary vascular resistance were relatively constant across the entire concentration range for sildenafil and bosentan. Overall, higher concentrations of imatinib and bosentan were not associated with increasing liver enzymes (serum glutamic oxaloacetic transaminases [SGOT]/serum glutamic-pyruvic transaminase [SGPT]). CONCLUSIONS: Population PKs of imatinib in patients with severe PAH were found comparable with those of patients with chronic myeloid leukemia. Imatinib was found effective regardless of the co-medications and showed intrinsic efficacy beyond merely elevating the concentrations of the co-medications due to DDIs. There was no evidence of increased risk of liver toxicity upon co-administration with bosentan.

Reduced sodium transport with nasal administration of the prostasin inhibitor camostat in subjects with cystic fibrosis.

BACKGROUND: Prostasin, a trypsin-like serine protease, is a channel-activating protease and major regulator of epithelial sodium channel-mediated sodium absorption. Its direct inhibition by camostat represents a potential approach to inhibiting sodium transport in cystic fibrosis (CF). METHODS: To determine whether a topical formulation of camostat represents an efficacious and tolerable approach to reducing Na+ transport in the CF airway, we conducted a two-part randomized, double-blind, placebo-controlled, crossover, ascending single-dose study to evaluate the pharmacodynamics, safety, and pharmacokinetics of camostat administered through a nasal spray pump in subjects with CF. Nasal potential difference (PD) was measured before and after treatment, and safety and pharmacokinetics were assessed by a standardized approach. RESULTS: In part 1, nine subjects were enrolled, and six completed crossover dosing at the maximally tolerated dose. The change in maximal (most polarizing) basal PD 2 h following administration of camostat was +13.1 mV (1.6-mg dose group) compared with -8.6 mV following placebo (P<.005). Intrasubject change in Ringer and amiloride-sensitive PDs exhibited similar and consistent responses. Bayesian analysis in an additional six subjects in part 2 estimated a dose of 18 µg/mL to provide 50% of the maximum effect. There was no significant change in chloride transport or total nasal symptom score, nasal examination rating, and laboratory parameters. CONCLUSIONS: This study establishes the proof of concept that a reduction in sodium transport in the human CF airway can be achieved through inhibition of prostasin activity, identifying a potential therapeutic target in the disease. TRIAL REGISTRATION: ClinicalTrials.gov; No.: NCT00506792; URL: www.clinicaltrials.gov.

A novel model-based approach for dose determination of glycopyrronium bromide in COPD.

BACKGROUND: Glycopyrronium bromide (NVA237) is an inhaled long-acting muscarinic antagonist in development for treatment of COPD. This study compared the efficacy and safety of once-daily (OD) and twice-daily (BID) glycopyrronium bromide regimens, using a novel model-based approach, in patients with moderate-to-severe COPD. METHODS: Double-blind, randomized, dose-finding trial with an eight-treatment, two-period, balanced incomplete block design. Patients (smoking history ≥10 pack-years, post-bronchodilator FEV1 ≥30% and <80% predicted, FEV1/FVC <0.7) were randomized to one of 16 independent sequences for 28 days. Primary endpoint: mean trough FEV1 at Day 28. RESULTS: 385 patients (mean age 61.2 years; mean post-bronchodilator FEV1 53% predicted) were randomized; 88.6% completed. All OD and BID dosing regimens produced dose-dependent bronchodilation; at Day 28, increases in mean trough FEV1 versus placebo were statistically significant for all regimens, ranging from 51 mL (glycopyrronium bromide 12.5 µg OD) to 160 mL (glycopyrronium bromide 50 µg BID). Pharmacodynamic steady-state was reached by Day 7. There was a small separation (≤37 mL) between BID and OD dose-response curves for mean trough FEV1 at steady-state in favour of BID dosing. Over 24 hours, separation between OD and BID regimens was even smaller (FEV1 AUC0-24h maximum difference for equivalent daily dose regimens: 8 mL). Dose-response results for FEV1 at 12 hours, FEV1 AUC0-12h and FEV1 AUC0-4h at steady-state showed OD regimens provided greater improvement over placebo than BID regimens for total daily doses of 25 µg, 50 µg and 100 µg, while the reverse was true for OD versus BID regimens from 12-24 hours. The 12.5 µg BID dose produced a marginally higher improvement in trough FEV1 versus placebo than 50 µg OD, however, the response at 12 hours over placebo was suboptimal (74 mL). Glycopyrronium bromide was safe and well tolerated at all doses. CONCLUSIONS: Glycopyrronium bromide 50 µg OD provides significant bronchodilation over a 24 hour period, and in terms of FEV1 AUC0-24h is not significantly different than the same total daily dose administered BID. Importantly, OD dosing may confer better patient adherence. The results are consistent with previous glycopyrronium bromide studies and support once-daily dosing of glycopyrronium bromide 50 µg in patients with moderate-to-severe COPD. TRIAL REGISTRATION: ClinicalTrials.gov: NCT01119950.

Pharmacokinetics of multiple inhaled NVA237 doses in patients with chronic obstructive pulmonary disease (COPD).

OBJECTIVE: NVA237 (glycopyrronium bromide) is a once-daily longacting muscarinic antagonist (LAMA) in development for the treatment of chronic obstructive pulmonary disease (COPD). This study investigated the pharmacokinetics (PK) of NVA237 following single and repeated once-daily inhalation in mild-tomoderate COPD patients. METHODS: In this double-blind, parallel-group study, COPD patients were randomized to a 14-day treatment with NVA237 (25, 50, 100 or 200 µg) or placebo. Plasma concentration-time profiles and urinary excretion of NVA237 were determined on Days 1 and 14. RESULTS: The median time to reach maximal plasma concentration (tmax) was 5 or 6.5 min postinhalation. At steady state (Day 14), total and maximum systemic exposure (AUC0-24, Cmax) to NVA237 and urinary excretion of unchanged drug (Ae0-24) was approximately dose proportional over the 50 - 200 µg dose range. The average exposure was 1.4- to 1.7- fold higher on Day 14 compared with Day 1. The mean terminal elimination half-life (t1/2) of NVA237 ranged between 13 and 22 h. Steady-state plasma concentrations were reached within 1 week of treatment. Renal clearance (CLR) was similar across doses both after single and repeated dosing, ranging between 17.4 and 20.6 l/h. Urinary excretion of NVA237 enantiomers ([3S,2R]- and [3R,2S]-stereoisomers) was similar with respect to the amount excreted within 24 h and the excretion rate. CONCLUSIONS: The pharmacokinetics of NVA237 were consistent between doses with limited systemic accumulation at steady state after repeated once-daily inhalation.

Efficient algorithms for optimal designs with correlated observations in pharmacokinetics and dose-finding studies.

Random effects models are widely used in population pharmacokinetics and dose-finding studies. However, when more than one observation is taken per patient, the presence of correlated observations (due to shared random effects and possibly residual serial correlation) usually makes the explicit determination of optimal designs difficult. In this article, we introduce a class of multiplicative algorithms to be able to handle correlated data and thus allow numerical calculation of optimal experimental designs in such situations. In particular, we demonstrate its application in a concrete example of a crossover dose-finding trial, as well as in a typical population pharmacokinetics example. Additionally, we derive a lower bound for the efficiency of any given design in this context, which allows us on the one hand to monitor the progress of the algorithm, and on the other hand to investigate the efficiency of a given design without knowing the optimal one. Finally, we extend the methodology such that it can be used to determine optimal designs if there exist some requirements regarding the minimal number of treatments for several (in some cases all) experimental conditions.

Characterization of the bronchodilatory dose response to indacaterol in patients with chronic obstructive pulmonary disease using model-based approaches.

BACKGROUND: Indacaterol is a once-daily long-acting inhaled ß2-agonist indicated for maintenance treatment of moderate-to-severe chronic obstructive pulmonary disease (COPD). The large inter-patient and inter-study variability in forced expiratory volume in 1 second (FEV1) with bronchodilators makes determination of optimal doses difficult in conventional dose-ranging studies. We considered alternative methods of analysis. METHODS: We utilized a novel modelling approach to provide a robust analysis of the bronchodilatory dose response to indacaterol. This involved pooled analysis of study-level data to characterize the bronchodilatory dose response, and nonlinear mixed-effects analysis of patient-level data to characterize the impact of baseline covariates. RESULTS: The study-level analysis pooled summary statistics for each steady-state visit in 11 placebo-controlled studies. These study-level summaries encompassed data from 7476 patients at indacaterol doses of 18.75-600 µg once daily, and showed that doses of 75 µg and above achieved clinically important improvements in predicted trough FEV1 response. Indacaterol 75 µg achieved 74% of the maximum effect on trough FEV1, and exceeded the midpoint of the 100-140 mL range that represents the minimal clinically important difference (MCID; ≥120 mL vs placebo), with a 90% probability that the mean improvement vs placebo exceeded the MCID. Indacaterol 150 µg achieved 85% of the model-predicted maximum effect on trough FEV1 and was numerically superior to all comparators (99.9% probability of exceeding MCID). Indacaterol 300 µg was the lowest dose that achieved the model-predicted maximum trough response.The patient-level analysis included data from 1835 patients from two dose-ranging studies of indacaterol 18.75-600 µg once daily. This analysis provided a characterization of dose response consistent with the study-level analysis, and demonstrated that disease severity, as captured by baseline FEV1, significantly affects the dose response, indicating that patients with more severe COPD require higher doses to achieve optimal bronchodilation. CONCLUSIONS: Comprehensive assessment of the bronchodilatory dose response of indacaterol in COPD patients provided a robust confirmation that 75 µg is the minimum effective dose, and that 150 and 300 µg are expected to provide optimal bronchodilation, particularly in patients with severe disease.

Omalizumab decreases IgE production in patients with allergic (IgE-mediated) asthma; PKPD analysis of a biomarker, total IgE.

WHAT IS ALREADY KNOWN ABOUT THIS SUBJECT: Omalizumab is a humanized anti-IgE monoclonal antibody that binds and captures circulating IgE, preventing interaction with receptors on mast cells and basophils, thereby interrupting the allergic cascade. It has a well-characterized efficacy and safety profile in patients with asthma. While omalizumab is known to reduce serum free IgE concentrations, effects on total IgE and IgE production are less well characterized. WHAT THIS STUDY ADDS: (i) Confirmation of prior hypotheses that IgE production can decrease with time when patients are given anti-IgE therapy; (ii) guidance on a biomarker, total IgE, which can be used to ascertain whether individual patients experience a change in their IgE production; and (iii) a way to assess whether patients' IgE production has been sufficiently down-regulated such that they may consider stopping anti-IgE therapy. AIM: To determine whether excessive IgE production by patients with atopic allergic asthma decreases with omalizumab therapy. METHODS: Omalizumab, free and total IgE data were obtained from an epidemiological study and six randomized, double-blind, placebo-controlled trials in patients with allergic asthma. The binding between omalizumab and IgE together with the production and elimination of IgE were modelled as previously, except that, in order to explain why total IgE was decreasing over a period of 5 years, the expression of IgE was allowed to change. RESULTS: The prior constant IgE production model failed to converge on the data once long-term observations were included, whereas models allowing IgE production to decrease fitted. A feedback model indicated that, on average, IgE production decreased by 54% per year. This model was further developed with covariate searches indicating clinically small but statistically significant effects of age, gender, body mass index and race on some parameters. Model predictions were checked internally and externally against 3-5 year data from paediatric and adult atopic asthmatic patients and externally against extensive total IgE data from a long-duration (>1 year) phase 1 study which was not used in the model building. CONCLUSIONS: A pharmacokinetic-pharmacodynamic model incorporating omalizumab-IgE binding and feedback for control of IgE production indicates that omalizumab reduces production of IgE. This raises the possibility that indefinite treatment may not be required, only for perhaps a few years. After the initial accumulation, total IgE should provide a means to monitor IgE production and guide individual treatment decisions.

Evaluation of biomarkers for pharmacological activity.

In recent years the cost of drug development has increased the demands on efficiency in the selection of suitable drug candidates. Biomarkers for efficacy and safety could be a plausible strategy to improve this selection process. In the present work, we focus on the study and evaluation of different physiological variables as biomarkers for pharmacological activity. We proposed three different approaches using multivariate and univariate techniques. We note that even though one could argue that the multivariate procedure is more powerful than the other alternatives, the univariate methods also offer a great flexibility to answer interesting scientific questions. The three approaches were used to analyze a crossover study involving an opioid antagonist.

Atomoxetine treatment of adults with ADHD and comorbid alcohol use disorders.

OBJECTIVE: Adults with attention-deficit/hyperactivity disorder (ADHD) have higher rates of alcohol and drug use disorders than adults without ADHD. The study aim was to determine if atomoxetine was superior to placebo in improving ADHD and alcohol use in recently abstinent adults with ADHD and comorbid alcohol use disorder. METHODS: Adults with DSM-IV diagnoses of ADHD and alcohol abuse and/or dependence were abstinent from alcohol at least 4 days (maximum 30 days) before study randomization. Participants received atomoxetine (25-100mg daily) or placebo for 12 weeks. ADHD symptoms were assessed using ADHD Investigator Symptom Rating Scale (AISRS) total score. Time-to-relapse to heavy alcohol use was analyzed using a 2-sided log-rank test based on Kaplan-Meier estimates and cumulative heavy drinking events over time were evaluated post hoc with recurrent-event analysis. RESULTS: Subjects received atomoxetine (n=72) or placebo (n=75) and 80 subjects completed the 12-week double-blind period (n=32 and 48, respectively). ADHD symptoms were significantly improved in the atomoxetine cohort compared to placebo (AISRS total score mean [S.D.], atomoxetine: -13.63 [11.35], P<.001; placebo: -8.31 [11.44], P<.001, difference: P=.007; effect size=0.48). No significant differences between treatment groups occurred in time-to-relapse of heavy drinking (P=.93). However, cumulative heavy drinking days were reduced 26% in atomoxetine-treated subjects versus placebo (event ratio=0.74, P=.023). There were no serious adverse events or specific drug-drug reactions related to current alcohol use. CONCLUSIONS: This 3-month, double-blind, placebo-controlled study of atomoxetine in adults with ADHD and comorbid alcohol use disorder demonstrates clinically significant ADHD improvement, and inconsistent effects on drinking behavior.

Accounting for variability in individual hierarchical clinical trial data.

Meta-analytical approaches have been extensively used to analyze medical data. In most cases, the data come from different studies or independent trials with similar characteristics. However, these methods can be applied in a broader sense. In this paper, we show how existing meta-analytic techniques can also be used as well when dealing with parameters estimated from individual hierarchical data. Specifically, we propose to apply statistical methods that account for the variances (and possibly covariances) of such measures. The estimated parameters together with their estimated variances can be incorporated into a general linear mixed model framework. We illustrate the methodology by using data from a first-in-man study and a simulated data set. The analysis was implemented with the SAS procedure MIXED and example code is offered.

Conditional mixed models with crossed random effects.

The analysis of continuous hierarchical data such as repeated measures or data from meta-analyses can be carried out by means of the linear mixed-effects model. However, in some situations this model, in its standard form, does pose computational problems. For example, when dealing with crossed random-effects models, the estimation of the variance components becomes a non-trivial task if only one observation is available for each cross-classified level. Pseudolikelihood ideas have been used in the context of binary data with standard generalized linear multilevel models. However, even in this case the problem of the estimation of the variance remains non-trivial. In this paper, we first propose a method to fit a crossed random-effects model with two levels and continuous outcomes, borrowing ideas from conditional linear mixed-effects model theory. We also propose a crossed random-effects model for binary data combining ideas of conditional logistic regression with pseudolikelihood estimation. We apply this method to a case study with data coming from the field of psychometrics and study a series of items (responses) crossed with participants. A simulation study assesses the operational characteristics of the method.

Antibiotic prescribing for acute cough: the effect of perceived patient demand.

BACKGROUND: GPs decide whether or not to prescribe antibiotics for acute cough. Apart from clinical signs and symptoms, non-medical reasons influence this decision as well. AIM: To obtain a valid estimate of the effect of perceived patient demand. DESIGN OF STUDY: Secondary analysis of cluster randomised controlled trial data. SETTING: Eighty-five Flemish GPs. METHOD: GPs completed a preprinted form with medical as well as non-medical information and their prescription for 20 consecutive adult patients consulting with acute cough in the periods February to April 2000 and 2001. The effect of perceived patient demand on antibiotic prescribing was estimated by performing alternating logistic regression analysis. A hierarchical backwards elimination procedure, described by Kleinbaum, was used. RESULTS: Seventy-two GPs participated, including 1448 patients eligible for analysis; 500 (34.5%) were prescribed an antibiotic and, according to the GP, 218 (15.1%) asked for an antibiotic. In cases of perceived patient demand antibiotics were prescribed significantly more often (odds ratio [OR] = 4.64, 95% confidence interval [CI] = 2.96 to 7.26). In the final model (n = 819; OR = 4.60, 2.59 to 8.17); Hosmer-Lemeshow goodness-of-fit P = 0.72), the effect of perceived patient demand for an antibiotic depended on the outcome of the lung auscultation. When patient demand was perceived, antibiotics were prescribed significantly more often only in the case of a normal lung auscultation or in the case of only one abnormal auscultatory finding (adjusted OR = 20.83, 95% CI = 8.86 to 48.99; adjusted OR = 4.79; 95% CI = 2.16 to 10.60, respectively). CONCLUSIONS: Perceived patient demand has a significant, independent and clinically relevant effect on antibiotic prescribing for acute cough with negative findings on the lung auscultation. Practice guidelines and interventions to optimise antibiotic prescribing have to take this effect into account.

Joint modelling of repeated measurements and event time: application to performance traits and survival of lambs bred in sub-humid tropics.

We considered the analysis of a study for Dorper, Red Maasai and crossbred lambs born over a period of 6 years at the Diani Estate, Kenya. The study was designed to compare survival and performance traits of genotypes with differing susceptibilities to helminthiasis. The available data include information on time to death and repeated measurements of body weight, packed cell volume (PCV) and faecal egg count (FEC) of the animals. In the paper, we consider joint modelling of the survival time and the repeated measurements. Such an approach allows to account for the possible association between the survival and repeated measurement processes. The advantages and limitations of the joint modelling are discussed and illustrated using the Diani Estate study data.

Applying concepts of generalizability theory on clinical trial data to investigate sources of variation and their impact on reliability.

This work aims at applying concepts of generalizability theory to data resulting from clinical trials. The focus is to study the sources of variance and their impact on the reliability and generalizability of a psychiatric measurement scale. The goal is to identify, measure, and thereby potentially find strategies to reduce the influence of these sources on the measurement in question for future trials. This approach was originally devised by Cronbach and his associates and is known as generalizability theory. This work shows how full modeling power in mixed models can be used to study generalizability using data from five double-blind randomized clinical trials, comparing the effects of risperidone to conventional antipsychotic agents for the treatment of chronic schizophrenia.

Prentice's approach and the meta-analytic paradigm: a reflection on the role of statistics in the evaluation of surrogate endpoints.

We put a perspective on the strengths and limitations of statistical methods for the evaluation of surrogate endpoints. Whereas using several trials overcomes some of the limitations of a single-trial framework (Prentice, 1989, Statistics in Medicine 8, 431-440), arguably the evaluation of surrogate endpoints can never be done using only statistical evidence but such evidence should be seen as but one component in a decision-making process that involves, among others, a number of clinical and biological considerations. We briefly present a hierarchical framework that incorporates ideas from Prentice's work and is uniformly applicable to different types of surrogate and true clinical outcomes.

Applying linear mixed models to estimate reliability in clinical trial data with repeated measurements.

Repeated measures are exploited to study reliability in the context of psychiatric health sciences. It is shown how test-retest reliability can be derived using linear mixed models when the scale is continuous or quasi-continuous. The advantage of this approach is that the full modeling power of mixed models can be used. Repeated measures with a different mean structure can be used to usefully study reliability, correction for covariate effects is possible, and a complicated variance-covariance structure between measurements is allowed. In case the variance structure reduces to a random intercept (compound symmetry), classical methods are recovered. With more complex variance structures (e.g., including random slopes of time and/or serial correlation), time-dependent reliability functions are obtained. The methodology is motivated by and applied to data from five double-blind randomized clinical trials comparing the effects of risperidone to conventional antipsychotic agents for the treatment of chronic schizophrenia. Model assumptions are investigated through residual plots and by investigating the effect of influential observations.

Statistical challenges in the evaluation of surrogate endpoints in randomized trials.

The validation of surrogate endpoints has been studied by Prentice, who presented a definition as well as a set of criteria that are equivalent if the surrogate and true endpoints are binary. Freedman et al. supplemented these criteria with the so-called proportion explained. Buyse and Molenberghs proposed to replace the proportion explained by two quantities: (1). the relative effect, linking the effect of treatment on both endpoints, and (2). the adjusted association, an individual-level measure of agreement between both endpoints. In a multiunit setting, these quantities can be generalized to a trial-level measure of surrogacy and an individual-level measure of surrogacy. In this paper, we argue that such a multiunit approach should be adopted because it overcomes difficulties that necessarily surround validation efforts based on a single trial. These difficulties are highlighted.