Rolling Day-Wise Mortality Prediction in Critically Ill Patients With AKI on CRRT Utilizing Machine Pressure Waveforms
Authors: Shehan Irteza Pranto, Joanna Yang, Joshua Lambert, Stuart L. Goldstein, Lili Chan, Girish N. Nadkarni, Tiago K. Colicchio, Javier A. Neyra, Jin Chen
Organizations: Department of Biomedical Informatics and Data Science, University of Alabama at Birmingham, AL, USA · Windreich Department of Artificial Intelligence and Human Health at the Icahn School of Medicine at Mount Sinai, New York, NY, USA · College of Nursing, University of Cincinnati, OH, USA · Department of Pediatrics, University of Cincinnati, OH, USA · Department of Medicine, Division of Nephrology at the Icahn School of Medicine at Mount Sinai, New York, NY, USA · Department of Health Economics and Systems Policy, Peter O’Donnell Jr. School of Public Health UT Southwestern medical center, Dallas, TX, USA · Division of Nephrology, Department of Medicine, University of Alabama at Birmingham, AL, USA
Critically ill patients with acute kidney injury (AKI) on continuous renal replacement therapy (CRRT) face high mortality, yet current risk assessment relies primarily on clinical parameters from electronic health records (EHR) and ignores minute-level circuit pressure waveforms generated by CRRT machines that track the extracorporeal circuit's interaction with the patient. Clinicians therefore cannot see deterioration as it develops. Risk is reassessed only when labs are drawn, while this continuous record is discarded because it is contaminated by shared-device records, non-physiological minutes, and sensor artifacts. To make the stream usable, we aligned machine records to charted therapy intervals to prevent cross-patient leakage, removed priming and downtime minutes, tuned denoising on a synthetic spike-injection benchmark, and masked unobserved intervals rather than imputing them. On this cleaned stream, we define a rolling day-wise task and a transformer-based stacked ensemble that late-fuses a window-reduced sequence transformer with classical models using circuit-instability features and clinical EHR variables. In a leak-safe benchmark on the multi-center CRRTnet cohort (976 patients, 4,585 treatment days), the machine-only model had the lowest standalone prognostic value (AUROC 0.625), followed by the EHR-only model (0.717). Integrating EHR and machine streams reached a one-day mortality AUROC of 0.766. SHAP attribution showed that circuit-instability descriptors raised the machine share of the top 15 combined-model features from 3 to 7 (20.0% to 46.7%), highlighting filter pressure, transmembrane pressure (TMP), and access-to-return difference (ARD). To our knowledge, this is the first patient-level mortality prediction incorporating CRRT machine data, turning a discarded bedside stream into a continuous risk signal.
Figures & tables
Fig. 1: Rolling day-wise prediction framework. The window advances one day at a time ( t−3,…,t ). Each treatment day defines a model-input window (dark), separated from the prediction horizon (green) by an excluded gap window (red) to guarantee causal, prospective risk estimates.
Variable
Survivors ( n=451 )
Non-surv. ( n=525 )
p
Age (Years)
57.0 [48.0–67.0]
63.0 [55.0–70.2]
< .001
Sex (Male)
205 (58.7%)
238 (63.3%)
.237
BMI, kg/m 2
30.1 [25.2–36.6]
29.7 [25.2–36.1]
.944
Charlson score
2.0 [1.0–3.0]
2.0 [1.0–4.0]
< .001
APACHE-II (part.)
9.0 [5.0–25.0]
13.0 [5.0–25.0]
.037
TABLE I: Patient baseline characteristics by in-hospital mortality.
Fig. 2: Signal-processing pipeline (top) and evaluation (bottom). The raw pressure stream is temporally aligned to clinical events, stripped of priming and downtime minutes, and denoised. The denoising operating point is optimized on a synthetic spike-injection benchmark (5% corruption): grid searches over detection parameters (window length, standard-score threshold) and smoothing configurations (neighbor count, mean/median aggregation) maximize spike-localization F1 and minimize reconstruction error (MAE, MSE).
Family
Features
/ch.
Ch.
N
Base set (133): distribution and within-day dynamics
slope, trend R2 , mean ∣Δ∣ , max jump, lag-1 autocorr., excursion frac.
6
7
42
Base subtotal
133
Enriched circuit-instability set (67)
Spectral
entropy, low/high band power, dominant period [ 27 ]
4
7
28
TABLE II: Taxonomy of engineered machine features. Pipelines use the full 200, reduced by mRMR to the selected machine budget (40 in the deployed model).
Fig. 3: Two-level stacked ensemble. (A) Each denoised treatment day yields a 144×30 window-reduced machine sequence (gap windows masked) and a 60-d mRMR clinical vector. (B) A Transformer encoder with a [CLS] token encodes the sequence, an MLP branch encodes the clinical vector, and a fusion head over the concatenated embeddings yields PDL . (C) Classical learners on the 100-d grouped-mRMR vector (40 machine + 60 clinical) emit P1,…,PK , ranked by validation AUROC. (D) A logistic-regression meta-learner, fit on validation, combines the four probabilities into Pstack , thresholded at the validation max- F1 point.
Fig. 4: Peak-detection performance on CRRT pressure waveforms. (a) Precision, recall, and F1 for the best outlier-detection model at each window length L ( F1 peaks at L=40 , F1=0.850 ). (b) Reconstruction error for smoothing variants at L=40 (mean N=2 is lowest, RMSE 9.14 ).
Fig. 5: Held-out test AUROC of the Windowed-Transformer Fusion Stack across the grouped-mRMR feature budget ( nmachine×nclinical ). The red box marks the validation-selected optimum, 40 machine ×60 clinical.
Fig. 6: Training and validation loss curves for the deployed combined fusion model (one-day mortality, class-weighted BCE, mean over five seeds with shaded range). The small train–validation gap indicates effective regularization; curves are truncated at the early-stopping patience limit.
Full (imbalanced) test
Balanced 1:1
Balanced 1:2
Model
AUROC
AUPRC
F1
Prec.
Rec.
AUROC
AUPRC
AUROC
AUPRC
Clinical-only
Clinical MLP
0.710
0.177
0.242
0.229
0.258
0.717
0.716
0.712
0.561
Machine-only
CNN1D a [ 32 ]
0.579
0.095
0.026
0.071
0.016
0.575
0.585
0.579
0.415
RNN (BiLSTM/GRU) a [ 33 ]
0.584
0.094
0.159
0.098
0.419
0.583
0.583
0.585
0.412
TABLE III: Deep-learning benchmark across clinical only, machine only and combined pipeline.
Full (imbalanced) test
Balanced 1:1
Balanced 1:2
Model
AUROC
AUPRC
F1
Prec.
Rec.
AUROC
AUPRC
AUROC
AUPRC
Clinical-only
LogReg
0.710
0.222
0.264
0.232
0.306
0.714
0.726
0.710
0.583
DecisionTree
0.617
0.117
0.215
0.177
0.274
0.614
0.608
0.614
0.444
RandomForest
0.711
0.233
0.239
0.173
0.387
0.713
0.731
0.709
0.589
GradBoost
0.717
0.209
0.277
0.203
0.435
0.720
0.721
0.716
0.575
TABLE IV: Classical benchmark across all three pipelines (one-day horizon; per-day, current denoised). Full imbalanced test with prevalence-balanced 1:1 and 1:2 resamples (mean of 100). Threshold = max- F1 on validation. Feature budget: clinical-only = 60 clinical, machine-only = 40 machine, combined = 40 machine + 60 clinical (grouped-mRMR).
TABLE V: Ablation of the deployed combined model (1-day; 40 machine × 60 clinical grouped-mRMR features), along two axes: signal preprocessing and sequence encoder. Held-out test AUROC; best in bold .
Combined model
Best rank
top-5
top-10
top-15
% top-15
Base
5
1
2
3
20.0
Enriched
2
1
5
7
46.7
TABLE VI: Machine features in the combined-model SHAP ranking: counts among the top-5/10/15 by mean ∣SHAP∣ , and the machine share of the top-15. Enriched = deployed optimal model (40 machine + 60 clinical grouped-mRMR).
Fig. 7: SHAP beeswarm (top-15 by mean ∣ SHAP ∣ ) for the Windowed-Transformer Fusion Stack, attributed via its Random-Forest tabular backbone. Red-boxed labels mark CRRT machine descriptors; positive values increase predicted mortality risk.
Postoperative acute kidney injury (AKI) after major non-cardiac surgery carries substantial morbidity, yet early intraoperative risk stratification remains difficult. In this retrospective cohort study, we propose SynerT, a waveform-only hybrid temporal backbone that combines a causal dilated TCN with a hierarchy of dilated recurrent layers to encode early intraoperative physiologic trajectories for AKI risk prediction. Building on SynerT, we further design two model variants that extend the backbone with structured clinical context: SynerT-MM, a late-fusion multimodal extension that integrates hemodynamic burden summaries and preoperative covariates, and SynerTStack, a leakage-safe stacked ensemble that combines cross-validated predictions from SynerT-MM with strong tabular baselines at the meta-learning stage. All models are evaluated under a strict leakage-aware framework on VitalDB, a high-fidelity perioperative database, with prediction restricted to information available within the first 60 intraoperative minutes. Among 2,413 waveform-usable cases (180 AKI-positive; 7.46% prevalence), SynerT fell well below strong structured-data baselines, demonstrating that waveform-only temporal modeling is insufficient under strict early constraints. SynerTMM recovered discrimination by incorporating hemodynamic burden summaries and preoperative covariates, and SynerT-Stack achieved the best overall performance across AUROC, AUPRC, and F1-max. Cross-fitted Platt recalibration substantially corrected calibration defects in both multimodal variants, and decision-curve analysis confirmed the recalibrated stacked model delivered the strongest net clinical benefit across low-to-intermediate thresholds.
Quang Minh Nguyen, Duc Minh Le, Ho Nhat Minh Nguyen +2
This study investigates a data-driven machine learning approach to predict membrane fouling in critically ill patients undergoing Continuous Renal Replacement Therapy (CRRT). Using time-series data from an ICU, 16 clinically selected features were identified to train predictive models. To ensure interpretability and enable reliable counterfactual analysis, the researchers adopted a tabular data approach rather than modeling temporal dependencies directly. Given the imbalance between fouling and non-fouling cases, the ADASYN oversampling technique was applied to improve minority class representation. Random Forest, XGBoost, and LightGBM models were tested, achieving balanced performance with 77.6% sensitivity and 96.3% specificity at a 10% rebalancing rate. Results remained robust across different forecasting horizons. Notably, the tabular approach outperformed LSTM recurrent neural networks, suggesting that explicit temporal modeling was not necessary for strong predictive performance. Feature selection further reduced the model to five key variables, improving simplicity and interpretability with minimal loss of accuracy. A Shapley value-based counterfactual analysis was applied to the best-performing model, successfully identifying minimal input changes capable of reversing fouling predictions. Overall, the findings support the viability of interpretable machine learning models for predicting membrane fouling during CRRT. The integration of prediction and counterfactual analysis offers practical clinical value, potentially guiding therapeutic adjustments to reduce fouling risk and improve patient management.
Alice Balboni, Luis Escobar, Andrea Manno +5
Department of Applied Science and Technology Politecnico di Torino Turin, Italy · Department of Inf. Eng., Computer Science and Mathematics University of L’Aquila L’Aquila, Italy · Dipartimento di Scienze della Salute University of Florence Florence, Italy +1
Accurate early prediction of Acute Kidney Injury (AKI) is critical for timely clinical intervention. However, existing deep learning models struggle with irregularly sampled data and suffer from the opaque "black-box" nature of sequential architectures, strictly limiting clinical trust. To address these challenges, we propose CT-Former, integrating continuous-time modeling with a Causal-Transformer. To handle data irregularity without biased artificial imputation, our framework utilizes a continuous-time state evolution mechanism to naturally track patient temporal trajectories. To resolve the black-box problem, our Causal-Attention module abandons uninterpretable hidden state aggregation. Instead, it generates a directed structural causal matrix to identify and trace the exact historical onset of severe physiological shocks. By establishing clear causal pathways between historical anomalies and current risk predictions, CT-Former provides native clinical interpretability. Training follows a decoupled two-stage protocol to optimize the causal-fusion process independently. Extensive experiments on the MIMIC-IV cohort (N=18,419) demonstrate that CT-Former significantly outperforms state-of-the-art baselines. The results confirm that our explicitly transparent architecture offers an accurate and trustworthy tool for clinical decision-making.
Weizhi Nie, Haolin Chen
School of Electrical and Information Engineering, Tianjin University