Optimal Hormone Replacement Therapy in Hypothyroidism - A Model Predictive Control Approach

Tobias M. Wolff, Johannes W. Dietrich, Matthias A. MüllerFrontiers in Endocrinology 2022 · DOI

TL;DR

This paper introduces a model predictive control (MPC) approach to optimize thyroid hormone replacement therapy for hypothyroidism. By extending a pituitary-thyroid feedback loop model to include oral drug intake, the authors demonstrate that MPC can directly determine optimal dosages, potentially replacing the current trial-and-error method. Their simulations suggest that a combined L-T3/L-T4 therapy is generally superior to L-T4 monotherapy, with nuances depending on genetic variants and drug intake frequency.

What Problem Does This Paper Address?

Hypothyroidism, a condition where the thyroid gland doesn't produce enough thyroid hormones, is typically treated with synthetic thyroid hormones. However, current treatment strategies face several challenges:

  • Dosage Determination: Finding the correct individual dosage of levothyroxine (L-T4) to normalize thyroid hormone levels can be difficult and time-consuming, often requiring frequent adjustments and potentially leading to temporary imbalances.
  • Therapy Suitability: It's unclear whether L-T4 monotherapy or a combination therapy with liothyronine (L-T3) and L-T4 is more effective for all patients, as some individuals continue to experience hypothyroid symptoms even with normalized TSH levels on L-T4 alone.
  • Optimal Intake Frequency: For combined therapies, the ideal frequency of L-T3/L-T4 intake to balance stable hormone levels with patient convenience is not well-established.

This paper aims to address these issues by developing a model-based approach to find optimal, personalized thyroid hormone replacement strategies.

Methods

The core of this work lies in the application of Model Predictive Control (MPC). The authors extend a previously established mathematical model of the pituitary-thyroid feedback loop. Key methodological contributions include:

  • Extended Pituitary-Thyroid Model: The model was enhanced to incorporate the oral intake, dissolution, and absorption of L-T3 and L-T4, using a two-compartment model adapted from existing literature. Bioavailability was considered for both euthyroid and hypothyroid individuals.
  • Model Predictive Controller (MPC): An MPC was designed to determine optimal dosages. MPC is a control strategy that predicts the system's future behavior and optimizes control actions over a defined horizon, allowing for the incorporation of constraints. In this context, the MPC aims to minimize deviations between simulated hormone concentrations (T3, T4, TSH) and their desired euthyroid setpoints.
  • Genetic Variant Modeling: The impact of specific genetic variants (polymorphisms in DIO1 and DIO2 genes) on enzyme activity was simulated by adjusting model parameters, allowing for an investigation into how these genetic differences might influence optimal therapy.
  • Intake Frequency Analysis: The model was used to simulate the effects of one, two, or three daily drug intakes on hormone concentration stability and patient convenience.

Datasets

The paper does not explicitly mention training on specific patient datasets. Instead, it relies on:

  • Existing Mathematical Models: The foundation is a previously developed mathematical model of the pituitary-thyroid feedback loop, extended with components for oral drug intake.
  • Literature Data: Parameter values for the model, including hormone half-lives and absorption rates, are derived from existing scientific literature and studies (e.g., (21) for bioavailability, (14) for drug intake models).
  • Generic Patient Profiles: The simulations are performed for a "generic hypothyroid patient" and also for simulated genetic variants. The model parameters for healthy individuals are based on well-known quantities and fitted to serum thyroid function tests of healthy individuals.

Key Insights and Ideas

  • MPC as a Replacement for Trial-and-Error: The most significant insight is that MPC can directly compute optimal dosages, potentially replacing the current, often lengthy and imprecise, trial-and-error approach to prescribing thyroid hormones. This offers a path towards more efficient and effective treatment.
  • Combined Therapy Nuances: While L-T3/L-T4 combined therapy generally appears slightly better for achieving target hormone levels, the benefit is not universal. Genetic variations can significantly alter the optimal strategy, highlighting the need for personalized medicine.
  • Genetic Influence on Therapy Choice:
    • For individuals with the AA genotype of the DIO1 polymorphism (rs2235544), L-T3/L-T4 combined therapy seems more beneficial.
  • Conversely, for the CC genotype of the DIO1 polymorphism, L-T4 monotherapy might be a better choice.
  • The DIO2 polymorphism (rs225014, CC genotype) also influences outcomes, with combined therapy showing advantages when aiming for the upper limit of the normal T3 range.
  • Intake Frequency Trade-off: For combined therapies, increasing the frequency of L-T3 intake (from once to twice daily) significantly reduces hormone fluctuations, leading to more stable levels. However, this comes at the cost of patient convenience, suggesting two daily intakes as a practical compromise.
  • High Initial Dosing: The simulations support the use of higher initial dosages followed by lower maintenance doses, which helps reach the euthyroid setpoint much faster than a "start low, go slow" approach, particularly for cardiac asymptomatic patients.

Key Results

  • General Superiority of Combined Therapy: In simulations without genetic variants, the L-T3/L-T4 combined therapy achieved slightly better hormone concentrations compared to L-T4 monotherapy.
  • Genetic Variant Impact:
    • For the simulated AA genotype of DIO1 (lower D1 activity), L-T3/L-T4 combined therapy restored the euthyroid setpoint, whereas L-T4 monotherapy showed a substantial offset.
  • For the simulated CC genotype of DIO1 (higher D1 activity), both therapies achieved the setpoint, with L-T4 monotherapy showing a better balance of T3/T4 concentrations in some scenarios.
  • For the CC genotype of DIO2, L-T3/L-T4 combined therapy resulted in better steady-state hormone concentrations when the T3 setpoint was at the upper limit of the healthy reference range.
  • Frequency of Intake:
    • Increasing intake frequency (from one to three times daily) reduced T3 fluctuations, especially for the L-T3/L-T4 combined therapy.
  • Two daily intakes of L-T3 were identified as a good trade-off between stable hormone concentrations and patient convenience.
  • Dosage Recommendations: The MPC provided specific optimal dosages, for example, a starting dose of 400 µg L-T4 followed by a maintenance dose of approximately 130 µg L-T4 for a generic hypothyroid patient.

Limitations and Future Work

  • Generic Patient Model: The current model represents a "generic" hypothyroid patient. Future work should focus on reliable individualization of the mathematical model by fitting parameters to individual patient data, leading to truly personalized treatment strategies.
  • Parameter Identification Challenges: Individualizing the complex model is not straightforward, as some parameters cannot be uniquely identified from limited individual hormone measurements. More frequent (dynamic) hormone data might help resolve this.
  • Commercial Availability of L-T3: The optimal L-T3 dosages determined by the MPC are often lower than currently available commercial L-T3 preparations, necessitating the availability of lower-dose formulations.
  • TRH Approximation: The model uses a constant TRH concentration, which is an approximation of the real pulsatile TRH course. While this approximation is deemed acceptable for T3 and T4 concentrations, further investigation might be warranted.
  • Cost Function Tuning: The paper notes that the specific hormone concentration profiles depend on the tuning of the MPC's weighting matrix (Q), suggesting that further exploration of different cost functions could yield varied outcomes.
  • Clinical Trials: The authors emphasize the need for clinical trials to translate these modeling results into practice and validate the MPC-derived strategies.

Summary updated

Abstract

In this paper, we address the problem of optimal thyroid hormone replacement strategy development for hypothyroid patients. This is challenging for the following reasons. First, it is difficult to determine the correct dosage leading to normalized serum thyroid hormone concentrations of a patient. Second, it remains unclear whether a levothyroxine <i>L</i>-<i>T</i><sub>4</sub>) monotherapy or a liothyronine/levothyroxine (<i>L</i>-<i>T</i><sub>3</sub>/<i>L</i>-<i>T</i><sub>4</sub>) combined therapy is more suitable to treat hypothyroidism. Third, the optimal intake frequency of <i>L</i>-<i>T</i><sub>3</sub>/<i>L</i>-<i>T</i><sub>4</sub> is unclear. We address these issues by extending a mathematical model of the pituitary-thyroid feedback loop to be able to consider an oral intake of <i>L</i>-<i>T</i><sub>3</sub>/<i>L</i>-<i>T</i><sub>4</sub>. A model predictive controller (MPC) is employed to determine optimal dosages with respect to the thyroid hormone concentrations for each type of therapy. The results indicate that the <i>L</i>-<i>T</i><sub>3</sub>/<i>L</i>-<i>T</i><sub>4</sub> combined therapy is slightly better (in terms of the achieved hormone concentrations) to treat hypothyroidism than the <i>L</i>-<i>T</i><sub>4</sub> monotherapy. In case of a specific genetic variant, namely genotype CC in polymorphism rs2235544 of gene <i>DIO1</i>, the simulation results suggest that the <i>L</i>-<i>T</i><sub>4</sub> monotherapy is better to treat hypothyroidism. In turn, when genotype AA is considered, the <i>L</i>-<i>T</i><sub>3</sub>/<i>L</i>-<i>T</i><sub>4</sub> combined therapy is better to treat hypothyroidism. Furthermore, when genotype CC of polymorphism rs225014 (also referred to as c.274A>G or p.Thr92Ala) in the <i>DIO2</i> gene is considered, the outcome of the <i>L</i>-<i>T</i><sub>3</sub>/<i>L</i>-<i>T</i><sub>4</sub> combined therapy is better in terms of the steady-state hormone concentrations (for a triiodothyronine setpoint at the upper limit of the reference range of healthy individuals). Finally, the results suggest that two daily intakes of <i>L</i>-<i>T</i><sub>3</sub> could be the best trade-off between stable hormone concentrations and inconveniences for the patient.
BibTeX
@inproceedings{wolff2022optimal,
  author = {Tobias M. Wolff and Johannes W. Dietrich and Matthias A. Müller},
  title = {{Optimal Hormone Replacement Therapy in Hypothyroidism - A Model Predictive Control Approach}},
  booktitle = {Frontiers in Endocrinology},
  year = {2022},
  doi = {10.3389/fendo.2022.884018},
}
Figure