Sutimlimab Pharmacokinetics and Pharmacodynamics in Patients with Cold Agglutinin Disease

Sutimlimab, a humanized monoclonal antibody targeting the classical complement pathway, is approved in the US, Japan, and EU for the treatment of hemolytic anemia in adults with cold agglutinin disease. The objectives of this study were to support dose selection for Phase 3 studies, assess dose recommendations, and establish the relationship between sutimlimab exposure and clinical outcome (hemoglobin [Hb] levels). Clinically meaningful biomarkers were graphically analyzed and the exposure-response relationship was proposed. The pharmacokinetic (PK) characteristics of sutimlimab were best described by a two-compartment model with parallel linear and non-linear clearance terms. Body weight was a significant covariate for the volume of distribution in the central compartment (Vc) and total body clearance of sutimlimab. Ethnicity (Japanese, non-Japanese) was a covariate on Vc and maximal non-linear clearance. There were no PK differences between healthy participants and patients. Following graphical exposure-response analysis for biomarkers, a pharmacokinetic-pharmacodynamic model was developed by integrating an indirect response/turnover model for Hb with a maximum effect (Emax) model, relating the Hb-elevating effect of sutimlimab to plasma exposure. Renal function and occurrence of blood transfusion were identified as covariates on Hb change from baseline. Simulations showed that Emax was attained with the approved dosing (6.5 g in patients <75 kg, and 7.5 g in patients {greater than or equal to}75 kg), independent of covariate characteristics, and provided adequate sutimlimab exposure to maximize effects on Hb, bilirubin, and total complement component C4 levels. A change in Hb from baseline at steady state of 2.2 g/dL was projected, consistent with Phase 3 study observations. Significance Statement The final validated population PK and PK/PD models confirm that the approved dosing regimen for sutimlimab (6.5 g in patients <75 kg, and 7.5 g in patients ≥75 kg) is sufficient, without the need for further dose adjustments in populations of patients with cold agglutinin disease.


Introduction (568/750 words)
Sutimlimab is a first-in-class, humanized immunoglobulin G 4 monoclonal antibody which targets the classical complement pathway (CP) by selectively inhibiting the complement component 1, s subcomponent (Röth et al., 2021;Bartko et al., 2018;Jäger et al., 2019). CAD is a rare type of autoimmune hemolytic anemia, mediated by persistent complement activation via the classical CP leading to hemolysis (Berentsen et al., 2016;Röth et al., 9 The Phase 3 CARDINAL (BIVV009-03; NCT03347396) and CADENZA (BIVV009-04; NCT03347422) trials examined the effect of sutimlimab in patients with CAD with or without a recent history of transfusion, respectively (Röth et al., 2021;Röth et al., 2022). In both trials, sutimlimab rapidly halted hemolysis and increased hemoglobin (Hb) levels. The activity of the classical CP was almost completely inhibited within 1 week following the initiation of sutimlimab treatment, with concomitant normalization of complement component 4 (C4) levels (Röth et al., 2021). The PK characteristics of sutimlimab in patients with CAD have not been published. The recommended dosage of sutimlimab for pivotal Phase 3 studies was 6.5 g for patients weighing between 39 kg and less than 75 kg, and 7.5 g for patients weighing greater than or equal to 75 kg. This was administered by intravenous infusion over 1-2 hours once weekly for the first two doses, followed by every 14 days thereafter (Enjaymo PI, Enjaymo SmPC).
The objectives of this study were to support dose selection for Phase 3 studies, to assess the effectiveness of the approved dose recommendation, to determine whether any further dose adjustments may be required, and to establish the relationship of sutimlimab exposure and clinical outcome, i.e., Hb levels. Population PK (popPK) and PK/PD models were developed to account for inter-subject variability and the effects of covariates on PK of sutimlimab and Hb, respectively. In addition, clinically meaningful biomarkers Hb, bilirubin and C4 values were graphically evaluated and the exposure-response relationship was proposed. Bilirubin is a recognized marker of extravascular hemolysis in CAD patients. C4 is the first soluble cleavable substrate of the C1 complex, the target of sutimlimab.

Study design and patient population
Data from the studies listed in Table 1 were included in the analyses. Detailed information on study design, dosing, and inclusion/exclusion criteria have been published previously (Röth This article has not been copyedited and formatted. The final version may differ from this version. JPET Fast Forward. Published on May 10, 2023 as DOI: 10.1124/jpet.122.001511 at ASPET Journals on July 23, 2023 jpet.aspetjournals.org Downloaded from et al., 2021;Röth et al., 2022;Jäger et al., 2019;Bartko et al., 2018;Müllbacher et al., 2017). These studies were conducted according to the International Conference on Harmonization Good Clinical Practice Guideline and the Declaration of Helsinki and were approved by local independent ethics committees or review boards. Participants provided written informed consent.

Model development
The popPK model was developed with data from studies BIVV001-01, BIVV009-02, BIVV009-03, and BIVV009-05, and further validated with a larger dataset as described in model validation. PK/PD model development included data from studies BIVV009-01 Part C and E, BIVV009-03 Part A and B, BIVV009-04 Part A, and was evaluated later using final data from Part B of studies BIVV009-03 and BIVV009-04.
(1) First structural and random models were identified including inter-individual variability (IIV) assessment.
(2) Following selection of the optimal structural PK or PK/PD model, trends in empirical bayes estimates versus categorical and continuous covariates were graphically tested first and identified for covariate model development and covariate analysis. A list of all covariates can be found in the supplementary material (Supp. Methods: Population PK model development, Population PK-PD model development).
Covariates were only considered on model parameters with identifiable IIV. The impact of identified covariates was then systematically evaluated using the forwardaddition (P = 0.01) and backward-deletion (P = 0.001) methods. Evaluation of the quality of both the PK and PK/PD models was based on likelihood of the data (objective function value) goodness-of-fit plots, η-shrinkage, quality criteria, visual predictive check (VPC), and simulations.
(3) The selected covariate model was then qualified by prediction-corrected visual predictive checks (pcVPCs) and bootstrap analysis.

PopPK model
The analysis population consisted of all evaluable participants, defined as those who had at least one post-dose sutimlimab concentration sample greater than the lower limit of quantitation (5 ng/mL) and an associated dosing and blood sampling record for the postdose sample.
Structural and random models were identified first, followed by covariate model development. Relationship of the following covariates was evaluated on volume parameters: age, body weight, sex, race, Japanese ethnicity, renal function by creatinine clearance (CLCR), and estimated glomerular filtration rate (eGFR), hepatic function (serum aspartate transaminase [AST], albumin), and disease status. Relationship of the following covariates was evaluated on clearance parameters: age, body weight, sex, race, Japanese ethnicity, CLCR, eGFR, measures of hepatic function, and disease status (Supp. Methods: Supp. Table S1). The impact of anti-drug antibody (ADA), diluted infusion, and occurrence of blood transfusion was graphically evaluated after availability of the complete dataset, as part of model evaluation. The final popPK model was then used to simulate expected profiles of potential populations of interest and further used to estimate exposure parameters of sutimlimab (minimum concentration [C min ] ss , maximum concentration [C max ] ss , area under the and dose group plots for observed and change (median, maximum, and 95th percentile) from baseline levels. An exploratory characterization of the exposure-response relationship was also attempted. A maximum effect (E max ) model was fitted to the change in Hb or C4 from baseline and a maximum inhibitory effect (I max ) model was fitted to the change in bilirubin from baseline versus time-matched observed C min,ss . Structural and random models were initially identified, followed by covariate model development (Supp. Methods: Population PK-PD model development). The covariates investigated were age, body weight, sex, race, ethnicity (Japanese/non-Japanese), blood transfusion, eGFR, CLCR, measures of hepatic function, i.e., albumin, AST, serum alanine transaminase, and bilirubin (Supp. Methods: Supp. Table S4). The impact of ADA and dilution of infusion was graphically evaluated. Further information on the popPK/PD model is provided in the supplementary material (Supp. Methods: Population PK-PD model development, Model evaluation).

Model validation
Maximum a Posteriori probability (MAP) Bayesian approach was applied to larger datasets to validate previously developed covariate models. The popPK and PK/PD full covariate model obtained in model development were applied to a dataset including data from prior analysis used for model development and additional data from completed studies (mainly This article has not been copyedited and formatted. The final version may differ from this version. BIVV009-03 and BIVV009-04), with prior population parameter estimates for the assessment of individual parameters and concentration predictions.
As the larger datasets were submitted to a MAP Bayesian analysis, potential new covariates were not included in the model and tested for statistical significance again. Instead, individual random subject effect estimates (η) were plotted versus covariates already investigated in previous analysis, such as ADA status, occurrence of blood transfusion, and administration of diluted/undiluted infusion. The approach is justified because the ηshrinkage was within an acceptable range.
Model evaluation was performed by standard goodness-of-fit plots, investigation of covariates, prediction-corrected VPC, and by quality criteria such as mean prediction error (MPE), root mean squared error (RMSE), and absolute average fold error (AAFE). Further information on the model evaluation is provided in the supplementary materials (Supp. Methods: Model evaluation). Trends in empirical bayes estimates versus categorical and continuous covariates were graphically tested (box and scatter plots). Absence of covariate effect was concluded if there was no major trend.

Dose selection
Concentration versus CP activity inhibition relationship was evaluated using data from BIVV001 Parts A -C and BIVV009-02 studies in healthy participants and patients. Individual serum concentrations of sutimlimab and CP activity were time-matched, and the PK/PD relationship was modeled using an inhibitory E max model as described below.
Where E 0 is the baseline, I max is the maximum inhibition, C is the concentration of sutimlimab, IC 50 is the concentration associated to 50% of the maximum effect, and H is the Hill factor (also referred as gamma, a parameter used to describe sigmoidicity). It was This article has not been copyedited and formatted. The final version may differ from this version. The popPK model was used to simulate the dose required in 97 participants (66 healthy subjects and 31 patients from two Phase 1 studies) to maintain concentrations above threshold concentration, when administered once weekly for 1 week and bi-weekly thereafter. The weekly regimen for the first few doses was proposed to achieve rapid increase in concentrations due to the expected steep nature of the PD response and rapidly saturating processes that cause non-linearities in PK. The threshold concentration in participants was justified based on most of the observed concentrations displaying nonlinearities below these levels, exacerbating risk for breakthroughs occurring rapidly given likely dosing deviations. As such, maintaining levels above threshold concentration would be expected to provide an additional concentration buffer for dosing deviations or unexpected variability in PK. Percent of participants with C min,ss above threshold concentration was estimated. The sutimlimab dose for pivotal Phase 3 studies was selected based on maximum percent of participants with estimated C min,ss above the threshold concentration.
Simulations were performed representing populations (including the various sources of IIV).
For the stochastic simulations, dose and covariates were sampled from the observed data considering potential correlations among them. In the deterministic simulations, covariates were kept at median values, except for the covariate under evaluation. Covariates were evaluated using the minimum and maximum values from the observed data. Each participant in the dataset was simulated 100 times. In total, 17600 virtual participants for PK and 7200 for Hb were simulated. The simulations were statistically summarized using median (defined by the 50th percentile of the simulated values) and 90% prediction interval (defined by the 5th and 95th percentile of the simulated values).

Dose-safety analysis
Sutimlimab doses per kg body weight were categorized by quartiles, and the rates of safety events described above were compared across the quartile categories. The same analysis was repeated with body weight quartiles.

Data accounting and baseline demographics
The final PK model parameters were estimated during model development using 2470 PK observations from 154 participants. The percentage of post-dose PK observations below the quantification limit (BLQ) was 7%. The final PK analysis dataset for model evaluation using

PopPK model
Several potential structural models were investigated (one or two compartments, mixed-   Table 3. Body weight and Japanese ethnicity were found to be statistically significant covariates for CL and V c , and Japanese ethnicity was found to be a significant covariate for V c and V max . In addition, age had a significant effect on V c (Table 3). After including age, body weight, and Japanese participant covariates, the following covariates did not have an additional effect on the PK of sutimlimab: sex, race, albumin, hepatic function, renal function, and disease status. The effect of ADA was graphically evaluated.
Evaluation of the PK model using all data showed good agreement between predictions and observations in the goodness-of-fit plots in general (Supp. Results: Supp. with an RMSE of 45.2% and 27.1% for predicted data based on population parameter estimates (PRED) and IPRED, respectively. AAFE was 1.37 and 1.17 for PRED and IPRED, respectively, and thus near the optimal value of 1 (Supp. Results: Supp. Table 5).
Box and scatter plots of the individual post hoc estimates of random effects versus body weight, age, ethnicity, race, sex, ADA status, disease status, occurrence of blood transfusion, infusion dilution, and measures of renal and hepatic function did not reveal marked correlation that would require correction of the covariate model (data not shown).
The mean changes of post hoc C min,ss C max,ss , and AUC ,ss at steady state ranged from 0.7 to 0.8 in patients weighing ≥75 kg, from 1.2 to 1.4 in Japanese patients and from 0.8 to 1.0 in elderly patients compared to the reference populations. The calculated 95% CIs overlap 1 in many instances ( Figure 1).

Graphical exploration of biomarkers
Hemoglobin and C4 levels increased, and bilirubin decreased after administration of sutimlimab in patients with CAD. After placebo treatment, there were no visible changes over time for Hb, bilirubin, and C4 ( Figure 2). The median observed changes from baseline at steady state following the approved regimen were 2.6 g/dL for Hb, 0.23 g/L for C4, and -1.7 mg/dL for bilirubin; and the median biomarker levels at steady state were 11.8 g/dL for Hb, 0.28 g/L for C4, and 0.801 mg/dL for bilirubin following 6.5 or 7.5 g of sutimlimab.
C min,ss and change in biomarkers were time-matched for characterization of exploratory PK/PD relationship. The estimated E max (Hb and C4) or I max (bilirubin) captured maximum observed effect on biomarker. The estimated C min,ss for half maximum effect on Hb and bilirubin (EC min for Hb, IC min for bilirubin) were 50.7 to 298 µg/mL with all observed C min,ss (Supp. Results: Supp. Fig. 6).
20 transfusion, infusion dilution, and measures of renal and hepatic function did not reveal marked correlation that would require correction of the covariate model (data not shown).

Dose selection
Near-maximal classical CP activity inhibition was observed for sutimlimab concentrations >20 µg/mL ( Figure 3). A steep concentration-effect relationship was observed for the inhibition of CP activity. Based on the inhibitory E max model, a 50% inhibition of CP activity was predicted at 6.2 µg/mL of sutimlimab, and that of a IC 90 was 15.5 µg/mL. The very low IC 50 , combined with a Hill parameter of 2.4 suggests a very steep concentration effect relationship and that sutimlimab concentrations above 100 µg/mL would be sufficient to maintain a near-maximal knockdown of CP activity and avoid non-linear PK. The final popPK model was used to simulate steady-state concentration in patients with CAD. Three dose levels were simulated for comparison: 5.5 g, 75 mg/kg, and the approved regimen of 6.5 g for body weight <75 kg, and 7.5 g for body weight ≥75 kg. The approved regimen of 6.5 g for body weight <75 kg, and 7.5 g for body weight ≥75 kg (6.5 g/7.5 g) was selected based on the predicted exposure being above 100 µg/mL in the maximum number of patients. The cut-off value of 75 kg was selected to coincide with the median body weight of 631 participants extracted from a US electronic medical record and claims database. This 6.5/7.5 g tiered dosing regimen was predicted to maintain target trough concentrations >100 μg/mL throughout the dosing period in approximately 94% of patients with CAD (Supp. Results: Supp. Table 7) to minimize the risk of breakthrough hemolysis while providing sufficient safety margins. The data from BIVV009-03 and BIVV009-04 studies showed only seven patients (7/66 = 10.6%) with trough levels below 100 μg/mL during the treatment period without dose interruption, which was consistent with the predicted values based on the popPK model, 6.2% (90% CI: 2.0% -12.0%).
This article has not been copyedited and formatted. The final version may differ from this version. The final popPK and PK/PD models were used to simulate exposures ( Figure 4) and Hb profiles ( Figure 5) at the approved dosing regimen for adult patients (6.5 g dose in patients who weigh <75 kg, and a 7.5 g dose in patients who weigh ≥75 kg). Doses were administered on Day 0, Day 7, and then every 14 days thereafter through Week 125.
Stochastic simulations of the evaluation of the impact of dose (6.5 g in patients <75 kg, and 7.5 g in patients ≥75 kg) stratified by minimum (39 kg) and maximum (112 kg) body weight showed the Hb change from baseline time profiles to be very similar ( Figure 6).
The impact of statistically significant PK and PD covariates on the typical changes in Hb from baseline at steady state, with the approved dosing regimen, was shown to be of 1.  Fig. 11).

Exposure-safety analysis
The proportions of patients with selected safety events by serum sutimlimab-exposure quartiles are provided in the supplementary materials (Supp. Results: Supp. Fig 12 and Supp. Fig. 13) for the AUC ,ss and C max,ss exposure metrics. Each quartile includes 16 to 17 participants. A trend is observed towards higher rates of GI disorders with an increased C max,ss , and general disorders and administration site reactions for AUC ,ss and C max,ss . The difference in percent of patients experiencing adverse effects is between 6% and 7% between the lowest and highest AUC ,ss or C max,ss quartiles for GI disorders, general disorders, and administration site discomfort. This difference is considered not relevant. In

Dose-safety analysis
The proportions of patients with selected safety events by dose and body weight quartiles are provided in the supplementary materials (Supp. Results: Supp. Fig. 14 and Supp. Fig.   15). For GI disorders, the difference in percent of patients experiencing adverse effects is about 5% between the lowest and highest dose quartiles. For vascular disorders, the difference in percent of patients experiencing adverse effects is about 7% between the lowest and highest weight quartiles. In general, there does not appear to be a clear trend between the other adverse events and dose or weight.

Discussion (1489/1500 words)
The PK of sutimlimab was best described by a two-compartment model with parallel linear and non-linear clearance terms (Michealis-Menten elimination kinetics). This model adequately described the non-linearity of sutimlimab PK especially visible at lower doses (0.3 to 60 mg/kg). This model also adequately described the PK of sutimlimab in patients with CAD and healthy participants.
Body weight and ethnicity (Japanese) were identified as covariates on the PK of sutimlimab, signifying the importance of demographic factors in sutimlimab PK. However, other covariates such as sex, race, albumin, hepatic function, renal function, and disease status did not have an additional impact on sutimlimab PK.
Separate residual error models were used for study BIVV009-03 due to the large difference in observed concentration between steady-state dosing (BIVV009-03) and single-or multiple-dose studies. Goodness-of-fit plots indicate population predictions to deviate from This article has not been copyedited and formatted. The final version may differ from this version. The changes of exposure parameters (i.e., C min,ss , C max,ss and AUC ,ss ) at steady state in patients weighing ≥ 75 kg, elderly patients (age > 65 years), and non-Japanese patients relative to the reference populations respectively ranged from 0.7 to 0.8, 0.8 to 1.0, and 1.2 to 1.4, respectively, altogether indicating a relatively limited impact of the previously identified covariates. Simulated concentration-time profiles using the final model showed an overlap in exposure following the 6.5 g and 7.5 g doses, indicating the cut-off was appropriately selected. Additional simulations also showed that exposures decreased with increasing body weight and that exposures were greater in Japanese participants versus non-Japanese participants. However, minimum sutimlimab concentration at steady state was greater than the target of 100 μg/mL to prevent hemolysis, suggesting that the approved dosing regimen (6.5 g in patients <75 kg, and 7.5 g in patients ≥75   Deterministic simulations showed that E max was attained independent of the PK (i.e., body weight, ethnicity, and age) and PD (i.e., renal function and occurrence of blood transfusion) covariate characteristics. The approved dose regimen (6.5 g for body weight <75 kg and 7.5 g for body weight ≥75 kg) was selected to maintain concentrations above 100 µg/mL in the maximum number of patients for near-maximal suppression of CP activity. The observed data from BIV009-03 and BIV009-04 studies showed that trough levels were maintained above 100 μg/mL during the treatment period without dose interruption in 90% of patients. Predicted change in Hb based on population PK/PD was maximum at the therapeutic weight tiered dose regimen (maximum Hb is more than 2 g/dL) including extreme body weight patients (39 and 112 kg).
The majority of the patients (51/66) from BIV009-03 and BIV009-04 studies were in the 6.5 g treated group. The dose level mean in mg/kg of the 6.5 g treatment group is 108.6 ± 18.1 mg/kg and that of 7.5 g treatment group is 89.7 ± 10.1 mg/kg. The mean of the dose in mg/kg overlaps between the two dose levels. The adverse events from BIV009-03 and BIV009-04 were graphically evaluated for correlation with dose (in mg/kg) or weight quartiles. No major trend was observed in adverse events with dose or body weight. In conclusion, no effect of dose (in mg/kg) or body weight on safety was observed.
In summary, the present analysis showed that the PK and PK/PD of sutimlimab can be accurately predicted using the respective models. This rigorous assessment of PK and PD data confirms the approved dosing regimen of sutimlimab, without the need for further dose adjustments in specific populations. Similar changes of Hb, bilirubin, and total C4 levels from baseline following 6.5 g or 7.5 g of sutimlimab and negligible changes in the placebo group were observed. Exploratory dose/exposure-safety analyses indicate well-tolerated administration with the approved dosing regimen.

Data sharing statement
Qualified researchers may request access to patient level data and related study documents including the clinical study report, study protocol with any amendments, blank case report form, statistical analysis plan, and dataset specifications. Patient level data will be anonymized and study documents will be redacted to protect the privacy of our trial participants. Further details on Sanofi's data sharing criteria, eligible studies, and process for requesting access can be found at: https://www.vivli.org/.  Individual post hoc estimates of C min,ss , C max,ss and AUC ,ss at steady state were calculated using the approved sutimlimab dosing regimen (6.5 g in red, 7.5 g in green). The solid circles represent the mean values, with error bars representing the 95% CI.  0 g sutimlimab is shown in blue, 6.5 g in red and 7.5 g in green. The shaded areas represent the 2.5% -97.5% percentiles of the respective doses.   shaded areas represent the range between 5th and 95th percentiles of the simulated values.

Disclosure of conflicts of interest
The dotted vertical lines represent the start or end of treatment period.

Fig. 6.
Relationship between Hb change from baseline at steady state by weight.
The vertical dotted lines indicate the approved dosing: 6.5 and 7.5 g; the horizontal line represents: 2 g/dL change in Hb from baseline.

Hb, hemoglobin.
This article has not been copyedited and formatted. The final version may differ from this version. The exposure-response (PK/PD) model was developed with data from patients with CAD from BIVV009-01 Part C and BIVV009-03 Part A.
This model was evaluated with additional final PD data from BIVV009-01 Part E, BIVV009-03 Part B, and BIVV009-04 Part A and B.
CAD, cold agglutinin disease; EOS, end of study; EOT, end of treatment; ET, early termination; LPO, last patient out; PD, pharmacodynamic; popPK, population PK; PK, pharmacokinetic.  The final PK analysis dataset included 196 participants (healthy participants and patients from all five studies). The number of participants from each study/cohort can be found in Table 1.
a The final PD dataset included 72 patients with CAD (6 participants from study BIVV009-01 Part C and E, and all participants from BIVV009-03 and BIVV009-04).
This article has not been copyedited and formatted. The final version may differ from this version.

Residual variability
This article has not been copyedited and formatted. The final version may differ from this version.   c Effect of baseline CLCR on Hb 0 (exponent) is relative to the median CLCR of 78.03 mL/min: Hb 0i = Hb 0 (CLCR i /78.03) 0.122

Population PK model development
The base model was characterized by the following expressions: where A 1 and A 2 are the amounts of drug in the central and peripheral compartments, C 1 and C 2 the corresponding concentrations; V 2 corresponds to V c , V 3 corresponds to V p . V max , K m , CL, V p , Q, and V p were fitted. At 0 h, A 1 = dose and A 2 = 0.
The base population PK model was assessed through inclusion of inter-individual variability (IIV) terms on selected parameters. The feasibility of modelling IIV was evaluated for clearance (CL), distributional clearance (Q), central volume of distribution (V), peripheral volume of distribution (V2), maximum non-linear clearance (V max ), and affinity of saturable clearance (K m ). A diagonal variance-covariance structure for the tested random effects was used at this stage of the analysis. The IIV terms were applied to the model parameters, assuming the multiplicative (proportional) form given by the expression: = , where denotes the value for the k th parameter for the i th subject, denotes the typical value for the k th parameter, and denotes the IIV in the k th parameter for the i th subject and is assumed to have a mean of 0 and variance of ω 2 .
An additive plus proportional residual error model as below was initially evaluated in the base models. Other forms of residual error models were tested to improve stability/fit, if deemed

Population PK-PD model development
An E max model related the Hb-increasing effect of sutimlimab to the sutimlimab plasma exposure at time t using the following equations: where Hb(t) is the Hb at time t, k in and k out are rate parameters related to Hb production/elimination, C(t) is the concentration of sutimlimab at time t, Ef C is the effect as a function of the individual predicted drug concentration units, Ef C,d describes the Hb-elevating effect of sutimlimab, Ef C,p the placebo effect, E max is the maximal increase in Hb, and EC 50 is the concentration required to reach 50% of Emax.
The turnover rate parameters were estimated:

Statistical models
The random effects of pop PK/PD included IIV. The distribution of the parameters was assumed log-normal, so that exponential models were used to account for IIV: P i = TVP ⋅ e ηi , where P i is the estimated parameter value for a given individual i, TVP is the typical population value of the parameter, η i describes the variation of individual i from the population estimate. In all cases, η is assumed to be normally distributed: η i ∼ N(0, ω 2 ), with inter-individual variance-covariance matrix Ω.
The statistical model for residual variability defines the nature of the interaction between the error and the predicted value to yield the observed value. The difference between the j th observed value (Y) in the i th individual and its respective prediction (F) was modeled with an additive error model: Y ij =F ij + ε i,j . The εi,j are independent, identically distributed statistical errors with a mean of 0 and a variance of σ 2 . The additive model for residual variability assumes that the variance of observations remains constant with the predicted values and the estimate is expressed as a standard deviation (σ).

Covariate analysis
Plots of IIV versus covariates defined in Supp.

Goodness-of-fit plots
Goodness-of-fit was graphically evaluated. The observed sutimlimab concentrations versus predicted concentrations, and observed Hb versus predicted Hb, were investigated to determine if the model described the data accurately. The descriptive performance of the models was also evaluated by calculation of normalized prediction distribution errors (NPDE). The NPDE should follow a (0, 1) distribution, and statistical tests were used to evaluate significant differences from the theoretical (0, 1) distribution.

Quality criteria
The measured (observed) concentrations (DV) and the corresponding predicted concentrations (PRED and IPRED) were evaluated in terms of mean prediction error (MPE; bias), root mean squared error; (RMSE; precision), and absolute average fold error (AAFE) using the following formulas: Where IPRED is the dividual predicted data based on individual empirical Bayes parameter estimates and PRED is the predicted data based on population parameter estimates (η = 0).
MPE and RMSE are also expressed as a percentage of the mean observed concentration value (DV):

Visual predictive check (VPC)
In the prediction corrected (pc) VPC, the observed and simulated data are normalized using the population prediction: PRED bin is the median of the population prediction of all observations in the respective bin. The same prediction-correction is applied for the observed measurements (y i,j ).

Exposure parameters
The C min,ss , C max,ss , and AUC ,ss at steady state were estimated for each subject included in this analysis considering dosing schedule and regimen as defined in studies 903 and 904.
Calculations were performed in R using the prior model structure and the individual post hoc estimates from NONMEM model. Steady state was assumed to be attained after 100 weeks.
AUC ,ss was calculated using the trapezoidal rule and C min,ss was the estimated sutimlimab concentration 1 hour prior to the next dose administration at steady state. C max,ss was the maximum concentration occurring in this interval at the end of infusion. The steady state C min,ss , C max,ss , and AUC ,ss ratios relative to the reference population were calculated by dividing the individual post hoc estimates of C min,ss , C max,ss , and AUC ,ss by the respective reference population mean C min,ss , C max,ss , or AUC ,ss . The reference populations were: patients weighing <75 kg, elderly patients (age >65 years) and non-Japanese patients.
The C min,ss , C max,ss , and AUC ,ss values were summarized by descriptive statistics. Boxplots were used to visualize C min,ss , C max,ss , and AUC ,ss versus dose (6.5 and 7.5 g), grouped by study and by previously identified covariates including occurrence of blood transfusion and excluding race and creatinine clearance. Forest plots trellised by dose group were used to visualize the mean and 95% CI of the change in C min,ss , C max,ss , and AUC ,ss at steady state relative to the reference population.

Handling of missing or erroneous data
Concentration samples missing corresponding dosing data were excluded from the analysis, as were samples with missing time or date information. Concentration samples below the lower limit of quantification (LLOQ) were treated as missing and excluded from the analysis. If more than 10% of the data were below the LLOQ, likelihood-based methods of imputation, (e.g., M3 likelihood imputation) were considered. When the value of a continuous demographic covariate was missing for an analysis subject, the median value from the rest of the population was used as the imputed value. Missing categorical covariates were flagged with an appropriate missing value code in the analysis dataset (i.e., -99), but grouped together with the most common covariate category during covariate model building.
Outliers were excluded from model-building, particularly during the covariate testing phase.
Prior to model-building, exploratory graphical analysis was used to identify unusual patterns and/or data points. Initial runs of the base population PK model were used to flag potential outlier values. Data observations for which the absolute value of the associated conditional weighted residual was greater than 4 (|CWRES| >4) were flagged as "questionable outliers".
These values did not create undue model destabilization or PK parameter changes and were not excluded from the analysis. The final population PK model was run using the entire dataset.
Missing variables were excluded by setting missing dependent variable (MDV) to 1. Biomarker values/sutimlimab concentrations below the LLOQ or which were flagged as missing were excluded from the analysis by setting MDV to 1. Duplicated records, Hb data with abnormally high (>25 g/dL) or low values (<1 g/dL), and post-dose concentrations without preceding dose records were excluded from the analysis by setting MDV to 1. Records with missing doses or missing treatment assignments were deleted from the dataset.

Supplementary Results Tables
Supp.