A ferritin prediction model to determine individual blood donation frequencies

Abstract

Background

Ferritin‐guided donation intervals are effective in reducing iron deficiency and low hemoglobin rates. However, many donors still present with low ferritin levels after multiple donations, which results in less donor availability. Model predictions of ferritin levels can help in setting donation frequencies that, proactively, keep ferritin levels above thresholds and donors available for donation.

Study Design and Methods

We created a physiology informed model to predict ferritin levels in whole blood donors. This model takes as input: baseline ferritin and hemoglobin, donation intervals, donated volume and height and weight of a donor. We used data from several studies to tune our model and validated ferritin predictions on routinely collected donation data from new donors in the Netherlands that started in 2021.

Results

The model predicted ferritin levels after five donations accurately, with 71% being within the range determined from the measurement uncertainty. Average ferritin predictions show a slight underestimation and accuracy varied by sex and age. We created tables of optimal donation frequencies given baseline ferritin levels and weight, which were shown to have the most influence on predicted ferritin levels after multiple whole‐blood donations.

Discussion

Personalized donation intervals offer a promising approach to shift from a reactive management of iron deficiency to proactively protect donor health and preserve long‐term donor availability. The healthy donor effect could explain the slight underestimation of predictions on average, and model accuracy may be further improved by, for instance, adding information on menstrual blood loss to account for sex‐ and age‐related differences.

Abbreviations

Hb

hemoglobin

HDI

highest density interval

MCMC

Markov Chain Monte Carlo

ME

mean error

ODE

ordinary differential equation

RMSE

root mean square error

1. INTRODUCTION

Blood donor eligibility is usually monitored by pre‐donation hemoglobin (Hb) measurement. But despite this Hb monitoring, iron deficiency is common among blood donors. ¹ , ² , ³ Therefore, to further safeguard whole‐blood donors from iron deficiency and from becoming anemic due to regular blood donations, a growing number of blood establishments have introduced monitoring of ferritin levels. ⁴ , ⁵ , ⁶ , ⁷ In the Netherlands, the implementation of ferritin‐guided donation intervals was shown to reduce the prevalence of iron deficiency in whole‐blood donors and reduce the number of Hb deferrals. ⁶ The flip‐side of this policy is that many donors are now being deferred because of low ferritin levels instead, which decreases donor availability and may result in concerns about their health and increase the drop‐out rate of these donors. ⁸ , ⁹ A logical next step is a more proactive policy that uses modeling to help maintain donors' ferritin at sufficient levels, protecting donors' health and enabling them to make regular blood donations.

We have shown previously that the decline in ferritin after whole blood donations is well predictable using a computational model that incorporates existing models on Hb synthesis with iron homeostasis. ¹⁰ In that study, the model was built on existing knowledge of iron dynamics without further parameter tuning, yet already 64% of predicted ferritin levels were contained within the range of the measurement uncertainty. Because this model has many parameters, while the data on ferritin and Hb are sparse, it is difficult to tune all parameters to create the best predictions for our specific use‐case: blood donors. Furthermore, the many parameters make interpretation of their influence on predictions ambiguous.

In this paper, we report an improvement on the previous model by Paalvast et al. ¹⁰ The new model is simplified but has biologically interpretable parameters which can be tuned to donation data. By being able to tune these parameters to data from blood donors we hypothesize that we can obtain better prediction accuracy while retaining the physiological dynamics that determine the interplay between red blood cell production and repeated blood donations. While our model also predicts Hb levels, the focus of this paper is on predicting ferritin levels. After all, by ensuring that long‐term ferritin levels stay above a certain threshold, we also minimize the risk of Hb deferral due to iron deficiency. We validate the prediction of ferritin levels on routinely collected blood donation data. Then, we can use our prediction model to determine optimal donation frequencies that keep donors' ferritin at adequate levels.

2. MATERIALS AND METHODS

2.1. Model description

The computational model that is proposed in this study is based on the model from Paalvast et al. ¹⁰ This model consists of the three‐compartment erythropoiesis model by Tetschke et al. ¹¹ adapted to Hb loss in blood donors and extended with an iron storage compartment to accommodate ferritin levels. Because the time duration between donations is quite long and thus the data on Hb and ferritin is sparse, we propose a simplified model with only two compartments. This significantly reduces the number of parameters and allows us to tune these parameters to data that is specific for a donor population.

The new model consists of two connected compartments: a Hb compartment and an iron storage compartment as shown in Figure 1. Like the previous model, the new model assumes that there is a personal baseline for Hb (${\mathrm{Hb}}_{\text{base}}$) and storage iron (${\text{iron}}_{\text{base}}$) such that a donor will always recover to their baseline at infinite recovery time and that donors are in equilibrium before their first donation. New with respect to the original model is that the connection between the two compartments is moderated by iron storage such that when iron storage becomes lower, less iron is transferred from the iron storage to the Hb compartment, leading to a longer Hb recovery time. In this way, the model reflects that donors with low ferritin drop in Hb with respect to their baseline. ¹²

FIGURE 1.

Visualization of the physiology informed model used to predict ferritin and Hb after blood donation.

The system of ordinary differential equations (ODEs) is given by

$$\frac{d\mathrm{Hb}\left(t\right)}{\mathit{dt}}=\mathit{\alpha f}\left(\text{iron}\left(t\right)\right)\left({\mathrm{Hb}}_{\text{base}}-\mathrm{Hb}\left(t\right)\right)$$

$$\frac{d\text{iron}\left(t\right)}{\mathit{dt}}=\beta \left({\text{iron}}_{\text{base}}-\text{iron}\left(\mathrm{t}\right)\right)-\gamma \frac{d\mathrm{Hb}\left(t\right)}{\mathit{dt}}.$$

In these equations $\mathrm{Hb}\left(t\right)$ represents the amount of iron in Hb and is calculated by multiplying a donors' Hb level by their blood volume (as calculated by Nadlers formula ¹³ ) and the amount of iron present per gram of Hb, which is 3.38 mg. Storage iron is denoted by $\text{iron}\left(t\right)$ and is calculated using the equation given by Cable et al. ¹⁴ and multiplying by a donor's body weight.

There are three parameters that control the recovery time of Hb and storage iron: $\alpha$ controls the Hb recovery time and has units of $t^{-1}$, $\beta$ controls the iron storage recovery time with units $t^{-1}$ and $\gamma$ is an additional factor that controls the conversion of storage iron to Hb‐bound iron and should be close to unity.

The function $f\left(\text{iron}\left(t\right)\right)$ determines the dynamics of iron in the storage transferred to Hb. While it is obvious that Hb production will be impaired in individuals with low iron storage, the exact relationship between iron storage and Hb is not precisely known. The function should go to zero for low iron storage and have an approximately linear relationship between log10(ferritin) and drop in Hb after multiple donations. ¹² We chose the function

$$\begin{array}{c}f\left(\text{iron}\left(t\right)\right)=\exp \left(\kappa \left(\frac{\text{iron}\left(t\right)}{728\mathrm{mg}}-1\right)\right),\end{array}$$

where 728 mg is the amount of iron storage for a ferritin level of 100 ng/mL (using the function by Cable et al. ¹⁴ ) and $\kappa$ will be determined from the data. Note that $1/\alpha$ should now be interpreted as the half‐recovery time of Hb at a ferritin level of 100 ng/mL.

For each individual donor, the system of ODEs is solved consecutively for each next visit with $t$, the days since the last visit, and the model output of the previous visits Hb and ferritin being the input of the next visit. At each donation, $\mathrm{Hb}\left(t\right)$ is assumed to drop instantaneously by the amount of blood collected with respect to a donor's blood volume.

2.2. Setting

In the Netherlands, ferritin is measured at intake, at every fifth donation, and when the previous ferritin level was ≤30 ng/mL. Ferritin levels between 15 and −30 ng/mL result in a 6‐month deferral, and ferritin levels below 15 ng/mL in a 12‐month deferral. In addition, ferritin may be measured if donors are repeatedly deferred for low hemoglobin. Ferritin in donors is measured on the Architect Ci8200 (Abbott Laboratories, Illinois).

2.3. Data

We used data from the KIND ¹⁵ and FIND+ ¹⁶ studies and routinely collected data from whole‐blood donors that started donating in 2021. The setup of the study is visualized in Figure 2.

FIGURE 2.

Flowchart of the study to visualize which data sources (rectangles) were used and analyses (ellipses) were done at different stages.

The KIND study contains Hb and ferritin measurements at relatively short intervals after a blood donation for “new” (<=2 prior donations) and “regular” (>10 donations) male donors. We used the KIND data from “new” donors to determine priors for the parameters of our model and use these priors in the next step where we fit the model to data from the FIND+ study.

The FIND+ study contains Hb and ferritin measurements at every donation from 300 new whole blood donors. ¹⁶ These donors were followed from the start of their donor career and donated at least twice. There are 199 female and 101 male donors in this cohort with a median age of 29 years for males and 25 years for females. See Table 1 in Paalvast et al. ¹⁰ for more data characteristics.

TABLE 1.

Donor characteristics of first time donors in the Netherlands that started in 2021 with at least two ferritin measurements before the end of 2024.

Characteristic	Females	Males
	(n = 13,965)	(n = 7824)
Age [years], median (IQR)	29 (21–46)	34 (25–49)
Blood volume [L], mean (SD)	4.3 (0.5)	5.7 (0.6)
Start log10(ferritin/(ng/mL)) [−], mean (SD)	1.6 (0.4)	2.1 (0.3)
Body weight [kg], mean (SD)	72 (13)	86 (13)
Start Hb [mmol/L], mean (SD)	8.5 (0.5)	9.5 (0.6)
Height [cm], mean (SD)	171 (7)	184 (7)
Time between donations [days], median (IQR)	158 (133–209)	89 (70–113)
# whole blood donations [−], median (IQR)	5 (2–7)	8 (6–10)
# plasma donations [−], median (IQR)	0 (0–0)	0 (0–0)
Total ferritin measurements	36,956	22,277
Ferritin at baseline ≤30 ng/mL (%)	39.8%	2.9%
Ferritin at 5th don. ≤ 30 ng/mL (%)	46.5%	35.7%

To validate the model that was tuned to the study data described above, we used routinely collected donation data from Sanquin in the Netherlands from blood donors that started donating in the year 2021 (n = 59,923). Only new donors were included that made at least one whole‐blood donation, had ferritin and Hb measured at intake and had at least one other ferritin measurement available. Eight donors that did not have their height or weight registered were removed, resulting in the selection of 21,789 new donors. Males and females were analyzed separately. The dataset includes 173,176 visits of which 132,910 resulted in whole‐blood donations and 16,846 resulted in plasma donations. The amount of red cell loss due to plasma donations was assumed to be 50 mL (10% of a whole‐blood donation). ¹⁷

2.4. Model fitting

We developed a Bayesian hierarchical model that allowed us to set informative priors on parameters $\alpha ,\beta ,\gamma ,\kappa$ from study data and determine baselines ($\mathrm{H}{\mathrm{b}}_{\text{base}}$ and $\mathrm{iro}{\mathrm{n}}_{\text{base}}$) for individual donors while accounting for the average of these parameters present in the population. Males and females were fitted separately.

The data collected from donations does not contain information on what happens in‐between donations. It is thus important to set informative priors on the fit parameters so as not to obtain unphysiological results. Therefore, we used the data from the KIND study to estimate reasonable priors on the parameters $\alpha ,\beta ,\gamma$, and $\kappa$ which are then further refined using data from the FIND+ study.

The model was optimized using Markov Chain Monte Carlo (MCMC) that maximizes the posterior given the likelihood of the measured Hb and ferritin levels from the KIND and FIND+ data. The likelihood for ferritin measurements was assumed to be a Normal distribution on log10(ferritin) with an uncertainty (determined by the biological and measurement variation) of 0.11 ng/mL. ¹⁰ The likelihood for Hb measurements was given by a Normal distribution for Hb levels with an assumed measurement uncertainty on Hb of 0.38 mmol/L. ¹⁸ In addition, for Hb levels below the threshold there is a correction as described in Pothast et al., ¹⁸ because the measurement may be repeated up to three times. The priors used are given in Supplementary Table 4.

We used numpyro ¹⁹ as a probabilistic framework with diffrax ²⁰ using Tsit5 ²¹ to solve the system of ODEs. The analysis code can be found at https://github.com/martp91/ferritin_model_donation_frequencies.

2.5. Model validation

We used routinely collected donation data as described above to determine the accuracy of our model. After fitting the model to the FIND+ data using MCMC, we computed the means of the resulting posterior samples of the model parameters and used them as follows. Using the first ferritin and Hb measurement of each donor as baselines together with the height and weight of a donor, we prospectively simulated future ferritin levels at each donation given the time between donations and the collected blood volume at each donation. As goodness of fit metrics of our predictions of log10(ferritin), we calculated the mean error (ME), root mean square error (RMSE), and determined the percentage of predictions within the 95% CI of the measurement uncertainty. These metrics are split for males and females and on the donation number. Furthermore, we studied differences between females below and above 45 years as a proxy for onset of menopause.

We specifically looked at the accuracy at the fifth donation because this is normally when the first measurement of ferritin is taken after the baseline ferritin measurement. This measurement is also the most unbiased, because it is not done on indication, that is, influenced by previous ferritin measurements.

2.6. Optimal donation frequencies

Using our model we calculated long‐term ferritin levels, defined as a donor's ferritin level after 10 donations, as a function of baseline ferritin level and donation frequency per year. After this many donations long‐term ferritin levels converge to a lower bound which is determined by the baseline ferritin level and donation frequency. We used the mean model parameters and the median donor parameters (weight, height and baseline Hb) for each sex. The optimal donation frequency is found by demanding that long‐term ferritin levels remain above a lower‐bound threshold which we set at 20 ng/mL, a value that may be further refined in the future.

2.7. Sensitivity analysis of parameter inputs on model output

It was already shown in Paalvast et al. ¹⁰ that the baseline ferritin is the strongest determinant of long‐term ferritin levels after whole‐blood donations and thus determines the optimal donation frequency. Here, we study the effect of changing other input parameters (height, weight, sex and baseline Hb) on model predicted ferritin levels at a donation frequency of three per year. For these simulations, each parameter of interest is set to its 10th and 90th percentile as present in our population (Table 1) while the other parameters are fixed to their respective median values.

3. RESULTS

3.1. Study population

Characteristics of the donor population included in this study are summarized in Table 1. We included 13,965 new female and 7824 new male donors who started donating in 2021 and had at least two ferritin measurements (including baseline) before the end of 2024. The median number of whole‐blood donation attempts was five for female donors and eight for male donors, with no plasma donations recorded.

In total the dataset contains 57,507 ferritin measurements. At baseline, 39.8% of female donors and 2.9% of male donors had ferritin ≤30 ng/mL. After five donations, these proportions increased to 46.5% for females and 35.7% for males.

3.2. Fit results of KIND and FIND+ data

Posterior parameter estimates for the KIND dataset are provided in Supplementary Table 5, and those for the FIND+ dataset in Supplementary Table 6. All MCMC runs showed good convergence (Supplementary Table 7). Figure 3 shows an example of hemoglobin and ferritin trajectories of a male FIND+ donor, together with posterior model predictions generated by the MCMC procedure.

FIGURE 3.

Example trajectories for Hb (top) and ferritin (bottom) of a male donor from the FIND+ study. The uncertainty bars denote 94% highest density interval (HDI) of the model prediction using all data points shown here.

3.3. Model accuracy on donation data

Regression metrics for log10(ferritin) stratified by sex and donation number are shown in Table 2. Overall, the model tends to predict lower ferritin levels than measured. For example, at the fifth donation, the ME for females is 0.048, which corresponds to 12% on the linear scale; for males, ME is 0.035 (8% on the linear scale).

TABLE 2.

Accuracy metrics for predicting log10[ferritin/(ng/mL)] on the dataset of new donors that started in 2021 from routinely collected donations in the Netherlands.

Sex	Donation number	ME (measurement − prediction)	RMSE	% within measurement range	N donors
Females	2–4	0.119 ± 0.003	0.264 ± 0.003	62 ± 1	5284
	5	0.048 ± 0.003	0.236 ± 0.002	66 ± 1	6019
	6–9	0.010 ± 0.003	0.251 ± 0.003	63 ± 1	5692
	10	0.071 ± 0.018	0.292 ± 0.012	53 ± 3	239
	>10	0.020 ± 0.021	0.269 ± 0.014	57 ± 4	162
Males	2–4	0.081 ± 0.010	0.238 ± 0.012	71 ± 2	497
	5	0.035 ± 0.002	0.191 ± 0.002	76 ± 1	5849
	6–9	0.019 ± 0.003	0.223 ± 0.003	69 ± 1	4409
	10	0.055 ± 0.006	0.230 ± 0.004	67 ± 1	1749
	>10	0.053 ± 0.006	0.243 ± 0.004	63 ± 1	1588

Note: ME = mean error, shows the average deviation of the model with respect to the measurement, closer to zero is better. RMSE = root mean square error shows the average spread of model predictions with respect to measurement (lower is better). The % within measurement range is the % of predictions that fall into the range as determined by the 95% CI of the measurement uncertainty. Uncertainties on the metrics are obtained by 1000 bootstrap resamples and denote $\pm$ SD.

Prediction variability is larger in females than in males, with RMSE values of 0.236 and 0.191, respectively, at the fifth donation. Accuracy declines when predicting farther into the future, reflected in higher RMSE values for later donation numbers. At the fifth donation, 76% of predictions for males and 66% for females fall within 95% of the measurement range.

3.4. Optimal donation frequency

Long‐term ferritin levels as a function of baseline ferritin and donation frequency are shown for females in Figure 4; the corresponding male pattern is similar, with slightly higher tolerable donation frequencies for males (Supplementary Figure 6). Table 3 lists the donation frequencies required to maintain a ferritin level of 20 ng/mL after 10 donations for both sexes. At comparable baseline ferritin levels, males can generally sustain marginally higher donation frequencies than females.

FIGURE 4.

Computed long‐term ferritin levels after 10 donations as a function of baseline ferritin and donation frequency. Shown are the results of a median female donor from our 2021 new donor dataset. Red (<15 ng/mL), light‐red (<20 ng/mL), yellow (20–30 ng/mL), light‐green (30–60 ng/mL), green (>60 ng/mL).

TABLE 3.

Recommended donation frequencies for baseline ferritin levels.

Baseline ferritin [ng/mL]		Donation frequency per year
Females	Males
<50	<50	1
<90	<70	2
<150	<125	3
<300	<175	4
≥300	≥175	5

Note: Thresholds for baseline ferritin are selected from Figure 4 so as to keep long‐term ferritin levels above 20 ng/mL. Shown are results for males and females with median height, weight, and baseline hemoglobin from the 2021 new donor dataset (see Table 1).

3.5. Sensitivity analysis of model predicted donation frequencies

Body weight is the most influential factor affecting ferritin predictions, after baseline ferritin and donation frequency. A 10‐kg decrease in body weight results in a predicted ferritin level that is 13% lower after five donations at a donation frequency of three per year. Conversely, a 10‐kg increase leads to a 12% higher predicted ferritin. Baseline Hb has a smaller, but noticeable effect. A 0.5 mmol/L decrease (or increase) corresponds to a 5% higher (or lower) predicted ferritin level. These effects are consistent across sexes, suggesting much of the observed difference between males and females is driven by body weight and baseline Hb.

Figure 5 shows the resulting optimal donation frequencies for females needed to maintain a long‐term ferritin level of 20 ng/mL as a function of baseline ferritin and body weight (with similar results for males in Supplementary Figure 7). Overall, donors with higher body weight can sustain more frequent donations at a given baseline ferritin level.

FIGURE 5.

Optimal donation frequency as determined by our model by demanding long‐term ferritin levels (after 10 donations) remain above 20 ng/mL as a function of baseline ferritin and body weight. Results are shown for a median female from our 2021 new donors dataset.

4. DISCUSSION

We present a physiology informed predictive model of iron dynamics following repeated whole blood donations. This model is tuned to Hb and ferritin measurements of dedicated study data and validated on routinely collected blood donation data from the Netherlands. Using only baseline ferritin, Hb, height and body weight as inputs, the model predicts ferritin levels after five donations within the 95% CI range of measurement uncertainty 71% of the time. We use the model to determine optimal donation frequencies that prevent whole blood donors from developing too low ferritin levels. These donation frequencies are determined primarily by a donor's baseline ferritin level, and to a lesser degree also by body weight, where individuals with greater body weight may tolerate slightly higher donation frequencies.

Females show considerably more variation of ferritin levels with respect to the model prediction than males. Short‐term fluctuation in ferritin can be expected for pre‐menopausal females because of menstrual blood loss. These fluctuations may also cause a misestimate of their baseline ferritin level, which may not be fully captured by a single measurement at intake.

Indeed, when splitting the data between age groups <45 versus ≥45 years as a proxy for menopausal status, we noticed that model predictions for ferritin are more accurate for older females (63% versus 70% within the range of measurement uncertainty for pre‐ and postmenopausal women, respectively).

The model predictions on average underestimate the measured ferritin levels slightly (about 10% on a linear scale) on the routinely collected donation dataset. A possible explanation is (self‐)selection of donors: donors with slower recovery are more likely to be deferred and less likely to present for subsequent donations, providing a bias for observing higher ferritin levels. However, this bias may also be present in the FIND+ study. Furthermore, there are subtle differences in donor characteristics between the FIND+ study and the routinely collected donation data. Donors from the FIND+ study are slightly younger (females 29 vs. 25 years and males 34 vs. 29 years) and have lower body weight (females 72 vs. 69 kg and males 86 vs. 82 kg), which may predispose to slower recovery of ferritin levels in this group, in turn leading to underestimation of the measurements in the routinely collected data used for validation purposes.

The underestimation of ferritin levels by model predictions is larger in females than in males. Since females are more likely to have lower baseline ferritin levels and thus recover more slowly in terms of Hb‐levels, they have an increased likelihood of deferral and dropout due to (self‐)selection. We intentionally did not correct for this bias because our proposed policy targets the full population of new donors—not only those who persist—thereby avoiding survivorship bias. This also means that there is room for improvement: some donors may sustain more frequent donations than others for reasons we have not yet taken into account here, like genetic factors and menstruation status. ²² , ²³

We provide tables of optimal donation frequencies and long‐term ferritin levels for the average whole‐blood donor in the Netherlands. While we expect that these results may also be extended to blood donor populations of other countries, extrapolating requires careful consideration of possible differences in setting. An important limitation of our results is that we are unaware if the donors in our data take iron supplements. In the Netherlands, at‐risk donors are currently not actively supplied with iron supplements (like in some other blood establishments ⁵ ), but it is possible that donors take supplements on their own initiative or are recommended by donor physicians to do so. ²⁴ In our model, it is theoretically possible to simulate the effect of iron supplementation, provided the amount and duration are known. Finally, ferritin measurements lack standardization and there are considerable differences between ferritin levels measured via different assays. ²⁵ , ²⁶

An important question that remains is how to set a functionally relevant threshold for ferritin. There is currently no consensus for this. ²⁷ , ²⁸ We specified a long‐term floor of 20 ng/mL, which is slightly higher than the existing 15 ng/mL policy threshold to account for biological variation. Naturally, this threshold can and should be updated pending more evidence. Moreover, the setting of these thresholds ultimately lies with blood establishments, as it reflects how they balance donor health against the number of donations collected.

Although baseline Hb is an input to the model, we did not evaluate future Hb predictions. A major deterrent to pursue this is formed by the bias in registered Hb due to repeated capillary measurements. ¹⁸ In the future, provided availability of data with less biased Hb measurements, we can determine the corresponding drop in Hb for low ferritin levels. ¹² In line with a proposal to mitigate the inherent randomness resulting from threshold‐based deferral, another approach could be determining personalized donation frequencies aimed at limiting the drop in Hb from baseline. ²⁹ , ³⁰

In summary, we present a physiology informed model that enables donor‐specific prediction of ferritin trajectories and allows us to develop personalized donation frequencies. This approach allows a shift from reactive management of iron deficiency to proactive, data‐driven donor care aimed at preserving both donor health and long‐term donor availability.

FUNDING INFORMATION

This work was supported by Stichting Sanquin Bloedvoorziening (Grant/Award Number: PPOC23‐09/L‐6512753).

CONFLICT OF INTEREST STATEMENT

The authors have disclosed no conflicts of interest.

Supporting information

Supplementary Table 4 Parameters for the (log)‐Normal priors used to fit the data of the KIND study.

Supplementary Table 5. Parameters as estimated from the KIND study which will be used as priors for the FIND+ data.

Supplementary Table 6. Posterior estimates as determined from the FIND+ data. The mean values are used in the final model. HDI = Highest Density Interval.

Supplementary Table 7. Convergence statistics of MCMC for FIND+ data. ess: effective sample size, r_hat compares the variance between chains and should be <1.01 for good convergence.

Supplementary Figure 6. Computed long‐term ferritin levels after 10 donations as a function of baseline ferritin and donation frequency. Shown are the results of a median male donors from our 2021 new donor dataset. Red (<15 ng/mL), light‐red (<20 ng/mL), yellow (20–30 ng/mL), light‐green (30–60 ng/mL), green (>60 ng/mL).

Supplementary Figure 7. Optimal donation frequency as determined by our model by demanding long‐term ferritin levels (after 10 donations) remain above 20 ng/mL as a function of baseline ferritin and body weight. Results are show for a median male from our 2021 new donors dataset.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.