top of page
Aerial Image of the Beach

Personalise Vitamin D.


Accepted Article

Huang ZH and You T. (2021) Personalise Vitamin D3 Using Physiologically-Based Pharmacokinetic Modelling. CPT: Pharmacometrics & Systems Pharmacology. PDF

Download model in R, training data and test data in the zip archive in Supporting Information



Why Vitamin D ?

Vitamin D is important

  • Vitamin D is essential for intestinal calcium absorption and bone health.

  • Vitamin D may play potential roles in the prevention of osteoporosis, cancer, diabetics, autoimmune disease and COVID-19 (Benskin, 2020)


Vitamin D deficiency is common all around the world

  • Target: serum 25(OH)D level ≥75 nmol/L (25(OH)D is an active metabolite of vitamin D)

       Severely deficient:            < 30 nmol/L           (FNB & IOM, 1997)

       Deficient:                           30 – 49 nmol/L      (IOM, 2011)

       Insufficient:                       50 – 74 nmol/L      (Holick, 2007)

       Target for prevention:      ≥ 75 nmol/L           (Holick, 2007)

       Danger of toxicity:            > 250 nmol/L

  • In North America, the prevalence of severe vitamin D deficiency increased to 10% in 2001-2006: the elderly, pregnant women, the black community and obese population accounted for a large proportion (Ganji et al, 2012)

  • In Europe, vitamin D deficiency in most countries were over 20%, except some Nordic countries: traditional diet of cod and cod liver in Nordic regions is speculated to explain such difference (Feldman et al, 2018)

  • In the UK, National Dietary and Nutrition Survey shows higher prevalence of hypovitaminosis D (marked by serum 25(OH)D < 40nmol/L) in the north of UK than the south, and the prevalence of hypovitaminosis D in most regions are higher than 30% in spring (Hyppönen et al, 2007)

    • Serum 25(OH)D was poor in the elderly

    • Low serum 25(OH)D has also been observed in UK adolescents

What are the main challenges?


Guidelines for daily vitamin D intake issued by different countries are different

  • US: Recommended Dietary Allowance (RDA) proposed by Institute of Medicine (IOM) (Ross et al, 2011)

    • Target levels: 25(OH)D ≥ 50 nmol/L

    • 600 IU/d for 1 to 70 years old

    • 800 IU/d for 71 years old and over

  • UK:  Reference Nutrient Intake (RNI) proposed by Scientific Advisory Committee on Nutrition (SACN, 2016) 

    • Target levels: 25(OH)D ≥ 25 nmol/L

    • 400 IU per day all year round for the general UK population, including pregnant and lactating women and people at increased risk of vitamin D deficiency


No model is accurate enough to help one effectively achieve sufficiency with the right dose and schedule

  • Many clinical trials have been performed all over the world, yet data not readily available

  • No model can accurately predict vitamin D pharmacokinetics in a person

    • Some models are poorly calibrated and unreliable: overfitted with many implicit assumptions not supported by evidence (Sawyer et al, 2015)

    • A nonlinear mixed effects model could only recapitulate limited doses and could not generate reliable prediction for an individual: the modelling attributes differences among individuals to random errors and do not attempt to understand the differences to make accurate predictions

What have we done?


We made high quality data readily available

  • We found over 100 trials from peer-reviewed publications as credible sources of information: the majority was randomised control trials to guarantee quality

  • We put in huge efforts to digitise the graphs to tabulate the values for modelling 

  • We filled in the missing metadata in line with with the best practice to curate the most comprehensive, up-to-date and high quality dataset that characterise vitamin D pharmacokinetics (PK)

We built a novel physiologically-based pharmacokinetics (PBPK) model that is truly predictive

  • Our data exploration highlighted baseline 25(OH)D serum levels were connected with PK

  • We only used a subset of data to identify the right structure and right parameters of a model

  • Our PBPK model most economically represented a simple hypothesis for this observation

  • This model accurately predicted 25(OH)D PK at doses far higher than the training set (i.e. >2500 µg)

Our work will help millions achieve vitamin D sufficiency

  • Our modelling made it possible to predict unique PK in an individual for the first time

  • A serological test in conjunction with our modelling may enable personalise vitamin D

    • Assess suitability of any dose regimen for helping an individual achieve optimal vitamin D

    • Estimate the time it takes to reach target 25(OH)D level in order to schedule follow up blood test

    • Predict 25(OH)D levels in real world situations that involve missing doses and drug holidays to ensure compliance

Want to access our data?


Data set we compiled are available for academic and commercial purposes.

Write us an email and let us know (* marks required information)

  • Your full name *

  • Company * | Department * | Job title *

  • Email *

  • Phone 

  • Why you need to access the data *

  • How you plan to use the data *

  • Will your activity contribute to The Open Project *


If your email meets all requirements, we will schedule an online meeting to discuss the way forward.

bottom of page