跳到正文
原文
Nature 健康护理主题·· 25 天前AI 评分31

肝配对交换决策支持建模:整合临床阈值与操作可行性

Decision-support modelling for liver paired exchange integrating clinical thresholds and operational feasibility

AI 导读

肝配对交换决策支持模型将ABO相容性、GRWR≥0.8、FRLV≥30%及操作可行性纳入整数线性规划,以最大化可行移植配对。验证采用20对供受者的合成队列,参数按韩国、印度和北美真实LPE/LDLT项目分布抽样。模型用PuLP/CBC求解并以IBM CPLEX验证,在供受者各至多匹配一次及物流复杂度上限下最大化移植数。

正文

Appenidx A. Simulation cohort generation methodology

To validate the optimisation model, a synthetic cohort of 20 donor-recipient pairs was generated using Python 3.13. Clinical parameters were sampled based on distributions from real-world LPE and LDLT programs in Korea, India, and North America:

Blood Group Distribution (from East Asian prevalence): A: 30%, B: 25%, AB: 10%, O: 35%

Graft-to-Recipient Weight Ratio (GRWR): Uniform distribution between 0.6 and 1.6; constraint ≥ 0.8

Future Remnant Liver Volume (FRLV): Uniform distribution between 25 and 50%; constraint ≥ 30%

Random Seed for Reproducibility: np.random.seed(42).

These parameters reflect published clinical thresholds to ensure anatomical and immunological feasibility.

Appendix B. Integer linear programming (ILP) formulation

Let \({x}_{ij}\) be a binary decision variable indicating whether donor \({\varvec{i}}\) is matched to recipient \({\varvec{j}}\).

Objective: Maximise total valid transplants:

$$ Maximise\,\,\sum\limits_{i \ne j} {x_{ij} } $$

(19)

Subject to:

  • for each donor (i), the donor can be assigned to at most one recipient. Each donor matched at most once:

$$\sum\limits_{j} {x_{ij} } \le 1\,\,\,\,\,\forall i$$

(20)

  • for each recipient (j), the recipient can receive a graft from at most one donor. Each recipient receives one graft at most:

$$\sum\limits_{i} {x_{ij} } \le 1\,\,\,\,\,\forall j $$

(21)

  • Match allowed only if:

    • ABO compatible

    • GRWRij ≥ 0.8

    • FRLVij ≥ 30%

Appendix C. Python implementation summary

The matching algorithm was implemented using the PuLP linear programming package. The compatibility graph was constructed by evaluating all donor-recipient pairs for clinical feasibility. The optimisation was solved using the built-in CBC solver and validated with IBM CPLEX Optimisation Studio for robustness and speed.

Appendix D. Visualisation output

The compatibility network was visualised using networkx and matplotlib, illustrating:

  • Valid donor-recipient matches.

  • ABO compatibility edges.

  • Constraints satisfaction.

Let \(I\) be the set of donors and \(J\) the set of recipients. We define a binary decision variable \({x}_{ij}\) that equals 1 if donor \(i\in I\) is assigned to recipient \(j\in J\), and 0 otherwise. The primary objective is to maximise the number of clinically feasible matches:

$$ {\text{max}}\mathop \sum \limits_{i \in I} \mathop \sum \limits_{j \in J} x_{ij} . $$

(22)

Clinical feasibility is encoded via pre-specified thresholds for ABO compatibility, graft-to-recipient weight ratio (GRWR), and future remnant liver volume (FRLV). Let \(\delta_{ij}^{ABO}\), \(\delta_{ij}^{GRWR}\), and \(\delta_{ij}^{FRLV}\) be 0–1 parameter indicating whether the pair \(i,j \) satisfies the corresponding criteria. The constraints are:

$$ \begin{aligned} x_{ij} \le & \delta_{ij}^{ABO} ,x_{ij} \le \delta_{ij}^{GRWR} ,x_{ij} \le \delta_{ij}^{FRLV} ,\;\forall i \in I,j \in J, \\ \mathop \sum \limits_{j \in J} x_{ij} \le & 1\;\forall i \in I,\mathop \sum \limits_{i \in I} x_{ij} \le 1\;\forall j \in J, \\ \end{aligned} $$

(23)

ensuring that each donor and each recipient participates in at most one exchange. To capture operational feasibility we also impose a global resource constraint

$$ \mathop \sum \limits_{i \in I} \mathop \sum \limits_{j \in J} \ell_{ij} x_{ij} \le L_{{{\text{max}}}} , $$

(24)

where \({{\ell}}_{ij}\) is an aggregate “logistical complexity” score for performing the exchange between donor \(i\) and recipient \(j\), and \({L}_{\text{m}\text{a}\text{x}}\) is the maximum capacity under the assumed operational scenario. The resulting formulation is a mixed-integer linear assignment model, solved using IBM ILOG CPLEX 22.1.

来源:Nature 健康护理主题 · nature.com