Runze Li

Clinical prediction from electronic health records (EHRs) is complicated by pervasive missingness and label scarcity, which make performance sensitive to the match between data conditions and pipeline choice. Choosing the best pipeline for a new incomplete dataset still requires costly trial-and-error. We cast this as an algorithm selection problem and address two bottlenecks—instance scarcity and distance quality—that have so far prevented meta-learning from reaching clinical settings. Graph neural networks offer diverse strategies (patient similarity networks, bipartite imputation graphs, attention-driven feature interaction), yet no single architecture dominates across missingness patterns, and selecting the best pipeline for a new dataset remains a trial-and-error approach. Formal algorithm selection could automate this choice but requires many characterized meta-instances—more than clinical settings typically provide. We propose two solutions: (1) constructive instance augmentation, applying controlled quality perturbations (MCAR and MNAR missingness injection, label trimming) to 20 base EHR datasets to expand the meta-knowledge base to 83 characterized meta-instances, each described by a 10-dimensional missingness fingerprint, without additional model training; and (2) dynamic-supervised metric learning, using differential evolution to optimize fingerprint feature weights so that static distances preserve method-performance similarity captured by dynamic fingerprints, which require model sweeps and are unavailable at deployment. Under base-dataset-level leave-one-dataset-out cross-validation over 21 pipelines, the resulting metric-learned kNN recommender attains the highest win rate (20.5%) among non-oracle strategies on the augmented store, selecting the correct pipeline more often than any fixed default. At deployment, the recommender needs only the 10-dimensional static fingerprint with pre-learned weights; no sweep data is required for new datasets. Cross-domain evaluation on 25 external subsets (colorectal cancer, kidney disease, MIMIC-IV) demonstrates framework modularity: when the fingerprint module is adapted (standard meta-features in place of the missingness-specific set), the recommender achieves regret of 0.025 (55% below random selection).

Background: Observational biomedical studies facilitate a new strategy for large-scale electronic health record (EHR) utilization to support precision medicine. However, data label inaccessibility is an increasingly important issue in clinical prediction, despite the use of synthetic and semisupervised learning from data. Little research has aimed to uncover the underlying graphical structure of EHRs. Objective: A network-based generative adversarial semisupervised method is proposed. The objective is to train clinical prediction models on label-deficient EHRs to achieve comparable learning performance to supervised methods. Methods: Three public data sets and one colorectal cancer data set gathered from the Second Affiliated Hospital of Zhejiang University were selected as benchmarks. The proposed models were trained on 5% to 25% labeled data and evaluated on classification metrics against conventional semisupervised and supervised methods. The data quality, model security, and memory scalability were also evaluated. Results: The proposed method for semisupervised classification outperforms related semisupervised methods under the same setup, with the average area under the receiver operating characteristics curve (AUC) reaching 0.945, 0.673, 0.611, and 0.588 for the four data sets, respectively, followed by graph-based semisupervised learning (0.450, 0.454, 0.425, and 0.5676, respectively) and label propagation (0.475,0.344, 0.440, and 0.477, respectively). The average classification AUCs with 10% labeled data were 0.929, 0.719, 0.652, and 0.650, respectively, comparable to that of the supervised learning methods logistic regression (0.601, 0.670, 0.731, and 0.710, respectively), support vector machines (0.733, 0.720, 0.720, and 0.721, respectively), and random forests (0.982, 0.750, 0.758, and 0.740, respectively). The concerns regarding the secondary use of data and data security are alleviated by realistic data synthesis and robust privacy preservation. Conclusions: Training clinical prediction models on label-deficient EHRs is indispensable in data-driven research. The proposed method has great potential to exploit the intrinsic structure of EHRs and achieve comparable learning performance to supervised methods.

BACKGROUND Observational biomedical studies facilitate a new strategy for large-scale electronic health record (EHR) utilization to support precision medicine. However, data label inaccessibility is an increasingly important issue in clinical prediction, despite the use of synthetic and semisupervised learning from data. Little research has aimed to uncover the underlying graphical structure of EHRs. OBJECTIVE A network-based generative adversarial semisupervised method is proposed. The objective is to train clinical prediction models on label-deficient EHRs to achieve comparable learning performance to supervised methods. METHODS Three public data sets and one colorectal cancer data set gathered from the Second Affiliated Hospital of Zhejiang University were selected as benchmarks. The proposed models were trained on 5% to 25% labeled data and evaluated on classification metrics against conventional semisupervised and supervised methods. The data quality, model security, and memory scalability were also evaluated. RESULTS The proposed method for semisupervised classification outperforms related semisupervised methods under the same setup, with the average area under the receiver operating characteristics curve (AUC) reaching 0.945, 0.673, 0.611, and 0.588 for the four data sets, respectively, followed by graph-based semisupervised learning (0.450, 0.454, 0.425, and 0.5676, respectively) and label propagation (0.475,0.344, 0.440, and 0.477, respectively). The average classification AUCs with 10% labeled data were 0.929, 0.719, 0.652, and 0.650, respectively, comparable to that of the supervised learning methods logistic regression (0.601, 0.670, 0.731, and 0.710, respectively), support vector machines (0.733, 0.720, 0.720, and 0.721, respectively), and random forests (0.982, 0.750, 0.758, and 0.740, respectively). The concerns regarding the secondary use of data and data security are alleviated by realistic data synthesis and robust privacy preservation. CONCLUSIONS Training clinical prediction models on label-deficient EHRs is indispensable in data-driven research. The proposed method has great potential to exploit the intrinsic structure of EHRs and achieve comparable learning performance to supervised methods.