Integrating early response biomarkers in pharmacokinetic models: A novel method to individualize the initial infliximab dose in patients with Crohn's disease

Abstract

The use of model‐informed precision dosing to personalize infliximab has been shown to improve both the acquisition of concentration targets and clinical outcomes during maintenance. Current iterations of infliximab pharmacokinetic models include time‐varying covariates of drug clearance, however, not accounting for the expected improvements in the covariates can lead to indiscriminate use of higher infliximab doses and imprecise drug exposure. The aim was to identify changes in the four biomarkers associated with infliximab clearance (Xiong et al. model) and determine if integration of these dynamic changes would improve model performance during induction and early maintenance. We analyzed two cohorts of children receiving infliximab for Crohn's Disease. The E _max method was used to assess time‐varying changes in covariates. Model performance (observed vs. predicted infliximab concentrations) was evaluated using median percentage error (bias) and median absolute percentage error (precision). The combined cohorts included 239 Crohn's disease patients. We found from baseline to dose 4, the maximum changes in weight, albumin, erythrocyte sedimentation rate, and neutrophil CD64 were 4.7%, +11.7%, −62.4%, and −26.5%, respectively. We also found the use of baseline covariates alone to forecast future trough concentration was inferior to the E _max time‐varying method with a significant improvement observed in bias (doses 2, 3, and 4) and precision (doses 2 and 4). The integration of the four time‐varying biomarkers of drug clearance with pharmacokinetic modeling improved the accuracy and precision of the predictions. This novel strategy may be key to improving drug exposure, minimizing indiscriminate dosing strategies, and reducing healthcare costs.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

The anti‐TNF biologics (infliximab and adalimumab) are the only biologics approved for children with moderate to severe Crohn's disease. Prior studies have shown significant variation in infliximab pharmacokinetics in children with Crohn's disease and all patients with active inflammation. With the discovery of blood biomarkers associated with infliximab clearance, there is renewed interest in incorporating biomarkers in precision dosing to inform dose optimization strategies at the start of therapy and following therapeutic drug monitoring.

• WHAT QUESTION DID THIS STUDY ADDRESS?

The primary objective was to identify the expected changes in four biomarkers (weight, serum albumin, erythrocyte sedimentation rate, and neutrophil CD64 expression) of infliximab clearance during early treatment and evaluate whether integration of the time‐varying covariates would provide more precise drug concentration predictions.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

We found bias and precision of predicting infliximab concentrations were significantly improved when the E _max time‐varying model was incorporated into a population pharmacokinetic model.

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

Our findings are highly relevant to clinicians as we present a direct approach to select the starting infliximab dose for the individual patient to improve drug exposure. We also provide clinical pharmacologists with a novel approach to incorporating time‐varying covariates in model‐informed precision dosing.

INTRODUCTION

Inflammatory bowel disease (IBD), including Crohn's disease (CD) and ulcerative colitis (UC), is a chronic inflammatory condition of the gastrointestinal tract with approximately one quarter of patients diagnosed before age 20. ¹ Both the incidence and prevalence of pediatric‐onset IBD have been increasing internationally in the twenty‐first century. ² Pediatric‐onset CD typically exhibits a more aggressive phenotype compared to adult‐onset CD, leading to a higher rate of intestinal complications. ³ Currently, anti‐TNF biologics (infliximab and adalimumab) are the only biologics approved by the United States Food and Drug Administration (FDA) and the European Medicines Agency (EMA) for the treatment of children (6–17 years old) with moderate to severe CD or UC. Despite the recent FDA/EMA approval of more novel advanced therapies for adults with IBD, the anti‐TNF biologics remain first‐line therapy for the management of moderate to severe pediatric CD in North America ⁴ and are second‐line for moderate to severe UC.

Routine use of proactive therapeutic drug monitoring (TDM) for children receiving anti‐TNF biologics has been used to achieve superior rates of clinical and biochemical remission, a decrease in the incidence of high‐titer antidrug antibodies (ADA), and improved drug durability. ⁵ , ⁶ , ⁷ Not only has TDM been a key advancement, but optimized induction dose selection and the attainment of early (week 6, dose 3) and post‐induction infliximab trough concentrations have been associated with improved early clinical and biochemical outcomes. ⁸ , ⁹ , ¹⁰

In order to better characterize infliximab pharmacokinetics (PK) in children, our team developed a population PK model based on real‐world data obtained from children and young adults with CD. ¹⁰ Subsequently, a multidisciplinary team created an electronic health record (EHR) integrated model‐informed precision dosing (MIPD) platform to standardize anti‐TNF dose optimization strategies during the induction and maintenance phases for both children and adults with IBD at our center. ¹¹ During the development of the population PK model and dosing platform, we discovered model accuracy improved when the five covariates of drug clearance (serum albumin, patient weight, erythrocyte sedimentation rate [ESR], ADA, and the neutrophil CD64 [nCD64] ratio) were included. ¹⁰ However, as this population PK model was developed from the collection of both induction and maintenance (early and late) infliximab concentrations, we anticipated there would be model imprecision during the induction phase given the dynamic changes in the covariates of drug clearance when a patient is rapidly healing (responding to treatment). Moreover, as pediatric patients often require higher starting doses to achieve therapeutic target concentrations, ¹² there is a critical need to develop dosing platforms and modeling strategies that account for the dynamic changes in infliximab clearance during the first year of therapy.

Time‐varying covariate modeling in pharmacometrics can predict and quantify the PK course over time utilizing biomarkers, disease phenotypes, or clinical symptoms. ¹³ To the best of our knowledge, the effect of dynamic changes in the covariates, reflecting early treatment response, on the accuracy of predictions and its integration as a standard procedure to optimize biologic (infliximab) exposure during the induction phase has not been described in the IBD literature. In this study, the primary objective was to identify the expected changes in the biomarkers of infliximab clearance and evaluate whether integration of predicted early response in the time‐varying covariates (biomarkers) in the induction phase of infliximab dosing would provide more precise drug concentration predictions during the early (dose 3) and the immediate post‐induction period (dose 4).

METHODS

Study population

Two cohorts of children and young adults receiving infliximab for CD were included in this investigation. The Targeting the Inflammatory Signature to Personalize Biologics in Pediatric IBD (REFINE) study was a multicenter, observational cohort that prospectively enrolled 78 children and young adults starting infliximab for CD at four medical centers from 2014 to 2019. The REFINE eligibility criteria have been reported in prior investigations. ¹⁰ , ¹⁴ During the REFINE study, longitudinal biomarkers of inflammation and infliximab concentrations were determined throughout the first year of infliximab therapy. The Accuracy of a PK Physician Decision Support Dashboard for Children receiving Infliximab for IBD (APPDASH) cohort included 161 children and young adults with CD who started infliximab at Cincinnati Children's Hospital Medical Center (CCHMC) from 2019 to 2021. Inclusion criteria for this retrospective cohort included patients 2–22 years old, male or female, with luminal CD who had received a minimum of four infliximab infusions at CCHMC and had at least one infliximab TDM performed prior to receiving a fourth infliximab infusion (dose 4). We excluded those with UC or IBD‐unspecified, those with internal penetrating CD or patients with missing infusion data. The EHR was queried to collect infliximab dosing (mg), infusion dates, start and stop times, patient weight, height, and all laboratory data associated with each infusion. With both cohorts, we recorded patient demographics, CD extent, and severity using the Paris Classification ¹⁵ and verified all past and current CD treatments.

Trough concentrations

In the REFINE cohort, the trough concentration was obtained at every infusion for the first year (including a post‐infusion peak after dose 1, 3, and 4). At a minimum, the APPDASH cohort had a trough concentration obtained at either dose 3 or dose 4. Infliximab and ADA concentrations for both cohorts were measured with the same electrochemiluminescence immunoassay (ECLIA; Esoterix, LabCorp Specialty Lab, Calabasas, CA). The lower limit of quantification for infliximab is 0.4 μg/mL and 22 ng/mL for ADA. Any value below the limit of infliximab quantification was removed from the analysis.

Biomarkers of infliximab clearance

The laboratory results of interest included serum albumin (g/dL), ESR (mm/h), and nCD64 (ratio).

Study outcomes

Clinical remission was defined as a weighted pediatric CD activity index of (wPCDAI) 12.5 and off prednisone at dose 4. Clinical remission was only determined in the REFINE cohort as there was no reliable method to assess for wPCDAI remission with the retrospective APPDASH cohort.

Statistical analysis

Continuous variables are represented as means with standard deviations (SD) or medians with 25–75% interquartile range (IQR) depending on data distribution. The Wilcoxon signed‐rank test was used to compare paired groups. The Friedman test, followed by Wilcoxon signed‐rank test with Bonferroni correction, was used to compare non‐parametric paired data across more than three groups. PK analysis and simulations were performed by nonlinear mixed‐effects modeling (NONMEM) software (version 7.5.1; ICON Development Solutions, Ellicott City, MD). R software version 4.1.0 (https://www.R‐project.org) was used for graphical and statistical analysis.

Estimate time‐varying changes in covariate data

We hypothesized that a mathematical model that accounted for improvements in the biomarkers of infliximab clearance would improve the prediction of infliximab trough concentrations (precision and accuracy) compared to a method of using the baseline covariates alone. Therefore, the stepwise approach was to describe the time‐varying changes in the four biomarkers of infliximab clearance and quantify these changes as the delta percentage from baseline to the specific timepoint of interest. First, we compared the percent change in each of the four covariates (albumin, weight, ESR, and nCD64) from baseline (pre‐treatment) to dose 3 (week 6) and from baseline to dose 4 (~week 14).

Second, in order to incorporate time‐varying changes in the covariates into the previously published population PK model, ¹⁰ we evaluated multiple models typically used to describe time‐varying clearance of antibody drugs, including linear, exponential, and E _max models. ¹⁶ Out of those models, the E _max model was determined to best describe covariate data changes over the course of treatment. The E _max equation is below:

$$\text{Covariate values}=\mathrm{E}0*\exp \left(\frac{E_{\max }*\left(\text{Dose Number}-1\right)}{\left(\mathrm{D}50-1\right)+\left(\text{Dose Number}-1\right)}\right)$$

In this equation, E0 is the baseline covariate value, E _max is the maximum increase or decrease, dose number is the infliximab dose number (dose 1, dose 2, dose 3, or dose 4) and D50 is the dose number when the change was 50%. In this analysis, D50 could not be estimated due to the lack of information on time‐course covariate changes between doses. Therefore, D50 was fixed to 1.1 to describe the most drastic covariate changes observed between dose 1 and dose 2 in this study. The E _max parameters were estimated with the maximum‐likelihood method using the lm4 R package (version 1.1–32). The random effect of between‐subject variability to E0 was considered in the estimation of the parameters while between‐subject variability E _max could not be estimated due to limited data. Non‐parametric bootstrapping was performed to evaluate the E _max model parameter estimates. ¹⁷ The original dataset was resampled with 1000 replicate data sets, and the estimated medians and 5th and 95th confidence intervals from the bootstrap analysis were compared to the final E _max model parameter estimates.

Evaluation of model prediction

The time‐varying parameters for each biomarker were then incorporated into the previously published population PK model ¹⁰ to account for the dynamic changes in covariate data from baseline during infliximab treatment. We assessed the effect the time‐varying changes in covariate data had on model prediction for the infliximab concentration as compared to the observed (actual) infliximab concentration. Prediction‐correct virtual predictive check (pcVPC) was performed to evaluate the predictive performance of the population PK model incorporating the time‐varying E _max model. ¹⁸ The 5th, 50th, and 95th percentiles of the observed data were visually compared to the corresponding percentiles of the simulated data (n = 1000). Additionally, the predictive performance of the infliximab trough concentration was evaluated using available clinical data (observed infliximab concentration and covariate data) for each dose by considering realistic clinical situations (i.e., clinical data availability). Bayesian estimation ¹⁹ , ²⁰ was used to predict the trough infliximab concentration if the observed infliximab concentration was available. Parameters from a previously published population PK model ¹⁰ served as Bayesian priors and individual PK profiles were fitted to the observed infliximab concentrations using maximum a posteriori Bayesian estimation in NONMEM to predict infliximab concentrations at the next dose. The bias and precision of the predictions were evaluated using the median percentage error (MPE) and median absolute percentage error (MAPE), respectively, as follows.

$$\text{Median percentage error}\left(\mathrm{MPE}\right)=\text{Median}\frac{C_{\text{pred}}-C_{\mathrm{obs}}}{C_{\mathrm{obs}}}\times 100$$

$$\begin{array}{c}\text{Median absolute percentage error}\left(\text{MAPE}\right)=\hfill \\ \text{Median}\left|\frac{C_{\text{pred}}-C_{\mathrm{obs}}}{C_{\mathrm{obs}}}\right|\times 100\hfill \end{array}$$

C _pred is the predicted infliximab trough concentration and C _obs is the observed infliximab trough concentration.

Ethical Considerations

The REFINE cohort study was approved by the Institutional Review Board (IRB) at CCHMC, Nationwide Children's Hospital, Medical College of Wisconsin, and Connecticut Children's Medical Center prior to the start of recruitment. The CCHMC IRB also approved the APPDASH study. The original de‐identified dataset is available upon request to the corresponding author.

RESULTS

Patients

In total, the REFINE and APPDASH cohorts included 239 patients with CD. The mean (standard deviation) age at the start of infliximab was 13.6 (3.7) years with 92.1% <18 years old, 40.6% were female, and 86.6% were white. The mean starting dose was 7.7 (2.3) mg/kg with additional demographic data presented in Table 1. There were notable differences in the mean age at diagnosis, the starting dose (mg/kg), and the patient's weight (kg) at the first dose between the subjects in the REFINE and APPDASH cohorts.

TABLE 1.

Patient demographics and baseline disease characteristics.

Characteristic	Combined cohorts (n = 239)	REFINE (n = 78)	APPDASH (n = 161)
Age at diagnosis, years	13.1 (3.9)	12.2 (3.9)	13.5 (3.8)*
Age at first infusion, years	13.6 (3.7)	13.1 (3.7)	13.9 (3.8)
Female, n (%)	97 (40.6%)	28 (35.9%)	69 (42.9%)
White race, n (%)	207 (86.6%)	72 (92.3%)	135 (83.9)
Infliximab starting dose, mg/kg	7.7 (2.3)	6.5 (1.8)	8.2 (2.3)***
Crohn's location (L1:L2:L3)	31:46:162	9:7:62	22:39:100
Crohn's behavior (B1:B2:B3:B4)	186:26:18:9	66:7:4:1	120:19:14:8
Perianal phenotype, n (%)	30 (12.6%)	12 (15.4%)	18 (11.2%)
Biomarkers at first infusion
Weight, kg	48.3 (20.5)	44.6 (20.7)	50.1 (20.2)*
Serum albumin, g/dL	3.3 (0.61)	3.3 (0.6)	3.3 (0.62)
Erythrocyte sedimentation rate, mm/h	26 (24)	25 (23)	27 (24)
Neutrophil CD64 activity ratio	7.2 (4.6)	7.5 (4.8)	6.8 (4.3)

Note: All values are listed as the mean (standard deviation) or number (n) with percentage (%).

Abbreviations: B1, inflammatory; B2, stricturing; B3, penetrating; B4, both stricturing and penetrating; L1, ileum only; L2, colon only; L3, ileocolonic.

*p < 0.05; ***p < 0.001.

Estimate time‐varying changes in covariate data

In a prior population PK study, Xiong et al. identified five biomarkers of infliximab clearance (patient weight, serum albumin, ESR, ADA, and nCD64) that improved the model performance. ¹⁰ Combining the data from both cohorts, we evaluated the observed percent improvement (delta) in four covariates (weight, albumin, ESR, and nCD64) from baseline to dose 3 and from baseline to dose 4. We also identified the percent improvement for additional key outcomes, including patient's (a) with a dose 3 trough concentration ≥18 μg/mL, (b) with a dose 4 trough concentration ≥5 μg/mL, and (c) those who achieved clinical remission at dose 4 (Table 2).

TABLE 2.

The observed improvement in the four covariates of drug clearance from dose 1–dose 3 or dose 1–dose 4 by outcome.

Timepoint	Outcome	Δ weight	Δ albumin	Δ ESR	Δ nCD64
Dose 1 to Dose 3	A. Dose 3 (all subjects)	6%	16%	−41%	−25%
Dose 1 to Dose 3	B. Dose 3 trough ≥18 μg/mL	9%	17%	−21%	−17%
Dose 1 to Dose 4	C. Dose 4 (all subjects)	8%	16%	−25%	−15%
Dose 1 to Dose 4	D. Dose 4 trough ≥5 μg/mL	10%	15%	−33%	−26%
Dose 1 to Dose 4	E. Dose 4 clinical remission	10%	17%	−25%	−36%
	Mean (standard deviation)	8.6% (1.7)	16% (0.84)	−29% (8)	−24% (8.4)

Note: Dose 3, week 6 infusion; dose 4, first maintenance infusion; clinical remission based on a weight pediatric Crohn's disease index < 12.5 and off steroids at dose 4. Clinical remission was assessed in the REFINE cohort (n = 78) only.

Abbreviations: ESR, erythrocyte sedimentation rate; nCD64, neutrophil CD64 ratio.

In a subsequent analysis, we sought to estimate the time‐varying changes in the four covariates using the E _max method. The E _max model allows for estimating the mean changes of covariate data after starting infliximab treatment and a method to incorporate the time‐vary changes into the population PK model as E _max functions. From baseline to dose 4, we found the E _max for weight was 0.046 (+4.7%), E _max for serum albumin was 0.110 (+11.7%), E _max for ESR was −0.978 (−62.4%), and the E _max for nCD64 was −0.362 (−26.5%, Figure 1). The bootstrap analysis demonstrated the stability of the E _max model (Table S1).

FIGURE 1.

Estimate of time‐varying changes in the four covariates of infliximab clearance. The E _max method was used to evaluate the percent change in (a) weight, (b) serum albumin (ALB), (c) the erythrocyte sedimentation rate (ESR), and (d) the neutrophil CD64 ratio (nCD64) over the first four doses (dose 1, dose 2, dose 3, and dose 4).

Given the high rate of response to infliximab induction and ensuing changes in the covariates during the induction and early maintenance periods, we simulated the infliximab PK profiles for a 47 kg (median body weight) patient receiving 5 or 10 mg/kg with the inclusion of the E _max calculated time‐varying covariates and compared these to simulations without considering dynamic covariate changes (calculations with baseline covariates only). Whether starting at 5 or 10 mg/kg, the infliximab trough concentrations were predicted to be higher at each timepoint when all four covariates (weight, albumin, ESR, and nCD64) with the predicted time‐varying changes were applied (Figure 2). We also found that the predicted infliximab trough concentrations were consistently predicted to be higher whether one, two, or all three covariates (albumin, ESR, and nCD64) were included (Figure S1). Interestingly, the inclusion of weight alone as a time‐varying covariate did not predict a higher infliximab concentration for either the 5 or 10 mg/kg simulations (Figure S2).

FIGURE 2.

Simulated infliximab pharmacokinetic profiles. The predicted concentration‐time curves were simulated for a 47 kg patient receiving (a) 5 mg/kg and (b) 10 mg/kg. The dashed line (‐ ‐ ‐) is the predicted concentration curve using the baseline covariates alone. The solid line is the predicted concentration curve using the time‐varying covariates with the E _max model.

Evaluation of model predictive performance

Now that we identified the mean change in covariates and performed dosing simulations, we sought to evaluate model precision and accuracy by assessing the median differences in the observed and predicted infliximab trough concentrations. Using only the baseline covariates and the Xiong et al. population model, ¹⁰ we found that the median predicted infliximab concentrations at dose 2, dose 3, and dose 4 were lower than the median observed concentrations (Figure 3). The pcVPC results also showed that the model tends to under‐predict infliximab concentrations using the baseline covariate data (Figure S3A). Incorporating time‐varying covariate changes with the E _max model, however, resulted in comparable infliximab median concentrations between the observed and predicted (there was no statistical difference at any timepoint). The improved predictive performance with the E _max model was also demonstrated by the respective pcVPC (Figure S3B).

FIGURE 3.

Observed and predicted infliximab trough concentrations using different modeling approaches. At each dose (2, 3, and 4), the observed median (solid line) and range of infliximab concentrations (red circles) were compared to the predicted median (range) infliximab concentrations using the four baseline (pre‐treatment) covariates only (blue circles) or the time‐varying covariates estimated with the E _max model (purple circles). NS, not significant; **p < 0.01; ***p < 0.001.

Finally, we sought to evaluate model predictive performance for the next infliximab trough under real‐world conditions. For example, to predict a dose 2 infliximab trough, the model must rely on the baseline (pre‐treatment) covariates alone. Similarly, to predict dose 3 infliximab trough, the model can rely on the observed covariates at dose 1 and/or dose 2 (TDM is not commonly performed at dose 2). Yet, to predict a dose 4 infliximab trough, the model can use both the covariates (at doses 1, 2, and 3) and the infliximab trough concentration at dose 3 (week 6). In fact, proactive TDM at dose 3 has become more common in real‐world clinical practice, especially in patients anticipated to have more rapid drug clearance. ⁹ Based on these real‐world scenarios, we assessed the model performance to predict the next infliximab trough concentration using the prior covariates alone and then using the combination of covariates and infliximab trough concentrations (if available) at the last infusion. We then assessed bias and precision to predict a dose 2 infliximab trough concentration using (a) the baseline covariates only and (b) with the E _max time‐varying covariates. We found that the bias and precision were improved when the E _max time‐varying model was used (Figure 4a). The comparisons of dose 3 trough concentration predictions were performed with the following conditions: (a) with baseline covariates only, (b) with E _max time‐varying covariates, and (c) the observed covariates at dose 2. We found the observed and predicted trough concentrations using the E _max method were similar (Figure 4b). We also found predictions using the baseline covariates alone to forecast the dose 3 trough concentration were inferior to the E _max time‐varying method with a significant improvement in bias and a mild (non‐significant) improvement in precision.

FIGURE 4.

Comparison of observed and predicted infliximab trough concentrations using the different modeling approaches and the overall predictive performance of each approach. For each dose, a box and whiskers plot were made for observed (red) and predicted infliximab concentrations (blue). Additionally, bias (%) and precision (%) were calculated at (a) dose 2, (b) dose 3 and (c) dose 4 using the listed scenarios. Bias (%), a measure of accuracy, and precision (%) were calculated and comparisons between the different modeling approaches were conducted. NS, not significant; *p < 0.05; **p < 0.01; ***p < 0.001.

We evaluated dose 4 trough concentration predictions with the following scenarios: (a) baseline covariates alone, (b) time‐varying covariates only (without TDM), and (c) the observed dose 3 covariates and dose 3 measured trough concentrations. We found that the use of baseline covariates alone significantly underestimated the trough concentration at dose 4 (Figure 4c). Moreover, we found that the bias and precision were significantly improved when the time‐varying covariate prediction model or Bayesian estimation using the observed covariate and dose 3 trough concentrations were used. The detailed results for evaluating the prediction performance are further summarized in Table 3.

TABLE 3.

Overall model predictive performance with and without time‐varying covariate analysis.

Dose	Scenario	Data included for trough prediction	Number of obs.	Observed concentration (μg/mL)	Predicted concentration (μg/mL)	Bias (%)	Precision (%)
2	a	Baseline covariates	63	22	12.8	−34.1	37.5
2	b	Time‐varying covariates	63	22	17.7	−12.2***	26.4*
3	a	Baseline covariates	131	19	10.7	−37.1	49.1
3	b	Time‐varying covariates	131	19	16.9	−5.2***	37.8
3	c	Covariates at dose 2	131	19	15	−12.3***	34.5**
4	a	Baseline covariates	94	7.1	4.9	−28.4	58.9
4	b	Time‐varying covariate	94	7.1	8.5	24.9***	48.5*
4	c	Covariates at dose 3 and infliximab trough at dose 3 (Bayesian)	94	7.1	8.3	22.4***	37.4*

Note: Concentrations are presented as median values. Statistical significance was determined by comparing scenario b (or scenario c) to scenario a (baseline covariates) for each of the three doses.

Abbreviation: Obs, observations.

*p < 0.05; **p < 0.01; ***p < 0.001.

DISCUSSION

While the FDA‐ and EMA‐approved dose for infliximab is 5 mg/kg, contemporary real‐world studies have shown induction doses for children with IBD often vary between 5 and 10 mg/kg. ¹⁰ , ¹² Therefore, given the variability in dose selection, the complexity of predicting infliximab clearance in patients with IBD even with PK models, increased costs associated with higher doses and the potential risks of excessive infliximab exposure, this study was designed to assess the changes in the biomarkers predictive of infliximab clearance during induction to better inform population PK models and individualize the starting dose of infliximab for patients with moderate to severe CD.

Currently, without regular access to decision support tools or PK modeling software, clinicians rely on a trial‐and‐error‐based approach for the initial dose selection and subsequent TDM to personalize infliximab maintenance regimens. More advanced strategies to guide dose selection or optimization include the use of covariates of infliximab clearance such as weight, serum albumin, presence of ADA, and inflammatory biomarkers such as ESR, nCD64, and CRP. ¹⁰ , ²¹ , ²² , ²³ Additional studies have identified several patient and disease‐specific predictors associated with more rapid clearance and include age <10 years old, weight <30 kg, a low body mass index and the extent and severity of gut injury. ¹⁰ , ¹⁴ , ²⁴ Therefore, it was imperative to develop a more novel strategy to individualize the starting dose of infliximab using biomarkers of drug clearance and replace the current trial‐and‐error scheme.

Selecting the initial dose based on pre‐treatment covariates of drug clearance may not vary significantly for a therapeutic compound with a short half‐life. However, it poses several challenges for a therapeutic with a long half‐life (such as infliximab that varies between 12.4 and 18.8 days ¹⁰ , ²¹ ) and when the PK is dependent on the disease process itself, including antigen burden and a leaky gut. ²⁵ We, therefore, considered more formal methods to assess improvements in the biomarkers associated with infliximab clearance with the goal to integrate these expected changes to the previously published PK model as a method to optimize the initial exposure, safety, and costs of care while also striving to minimize over‐exposure and delays in third‐party approval. In clinical and systems pharmacology, models with time‐varying covariates are commonly integrated into PK/pharmacodynamic models to account for the influence of drug treatment on disease progression or response over time. ¹³ The integration of time‐varying covariates was a logical next step in selecting the correct starting dose for advanced therapies given the complex interactions between disease severity, elevated inflammatory biomarkers, and rapid drug clearance (poor exposure) ²⁵ with acute severe ulcerative colitis serving as the prototypical trial‐and‐error example (not directly examined in this study). ²³ In fact, in a cross‐sectional study of patients with CD undergoing colonoscopy while receiving infliximab, we found significant increases in drug clearance in patients with active CD as compared to those with achieved endoscopic healing. ²⁶

For this investigation, model performance with the four time‐varying covariates was evaluated using the Xiong et al. population PK model as this is the only infliximab population PK model that includes the nCD64 as a covariate of clearance. ¹⁰ Not only have elevations in nCD64 been shown to alter infliximab clearance, our team has shown baseline elevations in nCD64 is a significant risk factor for primary nonresponse and the development of ADA. ¹⁴ , ²⁷ Moreover, while the Fasanmade et al. population model was recently found to have good accuracy when compared to seven other PK models, ²⁸ the authors concluded the performance differences were small, several models were deemed appropriate for clinical use and the effect of nCD64 on clearance could only be calculated using the Xiong et al. PK model.

Given the high response rate to infliximab induction (75–88.4%), ¹⁰ , ²⁹ , ³⁰ it was not unexpected that patient weight and serum albumin improved, and the inflammatory biomarkers (ESR and nCD64) decreased during the early induction period. Whether we used the delta improvement approach or the E _max method, we not only found the biomarkers improved as expected but found that the use of the E _max method to predict early (dose 3 and dose 4) infliximab concentrations was superior to the use of the Xiong et al. population PK model ¹⁰ alone (the approach of relying on baseline covariates only). Interestingly, utilizing singular covariates (such as weight only) to simulate trough concentrations was consistently shown to predict a lower infliximab exposure as compared to using a combination of the biomarkers to predict future infliximab concentrations. Therefore, the potential consequences of not considering the effect of early treatment response on infliximab PK covariates when selecting a starting dose is over‐estimating the dose (higher) needed to achieve a target concentration, which may delay third‐party drug approval and unnecessarily increase healthcare costs associated with requesting more drug than would be required to achieve the same effective target concentration. Finally, although early ADA has been reported during induction, we did not include ADA in our model expansion given the scarcity of ADA in our two real‐world cohorts.

This study benefited from the considerable number of children and young adults that were included and the substantial number of infliximab concentrations obtained using the same laboratory (Esoterix, LabCorp). We applied straightforward statistical approaches to identify the expected improvements in the biomarkers of drug clearance and utilized these identified improvements to inform our baseline population PK model. Except for the nCD64 test, all other biomarkers included in the model are easy to obtain in any clinical setting. Furthermore, we clearly demonstrated that the use of weight alone was insufficient to accurately predict infliximab exposure during induction.

The variability between the two cohorts (differences in the time to start infliximab and starting dose) reflects the transformation in infliximab dosing, especially in pediatric IBD, and the importance of identifying a more uniform approach to dosing infliximab. Study limitations included the inability to retrospectively track clinical outcomes in the APPDASH cohort and the lack of more objective measures, such as endoscopy, in both cohorts. While fecal calprotectin was obtained in the REFINE cohort for some participants, we did not look to define delta differences for this biomarker, given the missingness of fecal calprotectin data in both cohorts. Additionally, while the REFINE cohort had TDM (research‐only) performed throughout induction, TDM use for the APPDASH cohort was at the discretion of the clinician.

Although the use of dosing platforms for infliximab is not yet universal, ¹¹ , ³¹ there remains a limited number of advanced therapies approved for children. Several studies have indicated that employing the standard infliximab dosing of 5 mg/kg is linked to trough concentrations below the desired target concentrations. ¹⁰ , ¹² , ³² Therefore, studies such as these are imperative for clinicians to individualize the induction doses in children to improve short and long‐term outcomes. In fact, the use of predicting early covariate changes to guide infliximab starting doses is being investigated in the multicenter REMODEL‐CD clinical trial (NCT05660746). ³³ Additionally, the CAMEO trial (NCT05781152) is investigating the use of a precision dosing platform to dose optimize anti‐TNF therapy following TDM in children with CD.

In this study, the D50 parameter for the E _max model could not be estimated due to the lack of covariate data between the first and second doses. Covariate data was only available at the time of infusion, with no intermittent data collected between infusions. Therefore, the D50 was empirically determined to describe the observed covariate changes over the course of treatment (from dose 1 to dose 4). To evaluate the empirically determined value, we performed a sensitivity analysis, which indicated that higher D50 values resulted in the continued increase (weight and albumin) or decrease (ESR and nCD64) in the covariates after doses 3 and 4. As those contradict the observed data in this study, we decided to use 1.1 for D50. Furthermore, the between‐subject variability of E _max could not be estimated due to the limited data. For those limitations, further research is warranted to fully characterize time‐varying biomarker changes during the induction phase, which could verify the findings in this study.

In conclusion, based on the improved accuracy of predicting drug exposure by incorporating time‐varying covariates to the population PK model, our team is currently conducting a multicenter clinical trial (REMODEL‐CD) to evaluate whether precision induction dosing of infliximab informed by predictions of early covariate improvement will result in superior rates of intestinal healing as compared to conventional dosing. Furthermore, we anticipate the use of clinical decision support tools that incorporate advanced methods of personalized dosing, such as RoadMAB™ (a bedside PK dosing platform), ¹¹ will become standard in real‐world clinical practice to ensure sufficient drug exposure targets are met for all biologic therapies.

AUTHOR CONTRIBUTIONS

A.S., P.M., K.I., and T.M. wrote the manuscript. P.M., K.I., T.M., N.P., and A.A.V. designed the research. P.M., K.I., T.M., A.S., J.R., N.P., and A.A.V. performed the research. P.M., K.I., T.M., N.P., and A.A.V. analyzed the data.

FUNDING INFORMATION

This work was supported by the National Institute of Diabetes and Digestive and Kidney Diseases at the National Institutes of Health (grant number DK132408) and by the Cincinnati Children's Research Foundation.

CONFLICT OF INTEREST STATEMENT

N.P. is president of Medimatics, a company that provides consulting services on medical information systems. A.A.V. and P.M. are inventors of the RoadMAB™ precision dosing platform. P.M. is currently receiving drug‐only (in‐kind) support from Janssen Scientific Affairs, LLC. All other authors declared no competing interests for this work.

Supporting information

Data S1:

Samuels A, Irie K, Mizuno T, et al. Integrating early response biomarkers in pharmacokinetic models: A novel method to individualize the initial infliximab dose in patients with Crohn's disease. Clin Transl Sci. 2025;18:e70086. doi: 10.1111/cts.70086