Pith. sign in

REVIEW 3 major objections 5 minor 44 references

DeepEN, an offline reinforcement learning framework for ICU enteral nutrition, is estimated to cut mortality by 3.7 percentage points versus clinician practice.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

An offline RL policy trained on ICU feeding records recommends personalized enteral nutrition targets and reports an estimated 18.8% vs 22.5% mortality advantage over clinicians.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection First RL-for-enteral-nutrition framework, thoughtfully designed; but the headline 3.7-point mortality gap is a circular relabeling of returns that already contain the terminal survival reward — don't take it at face value. the 3 major comments →

arxiv 2510.08350 v3 pith:EWPVYKQQ submitted 2025-10-09 cs.LG cs.AI

DeepEN: A Deep Reinforcement Learning Framework for Personalized Enteral Nutrition in Critical Care

Daniel Jason Tan , Jiayang Chen , Dilruk Perera , Kay Choong See , Mengling Feng This is my paper
classification cs.LG cs.AI
keywords reinforcement learningoffline reinforcement learningenteral nutritioncritical carepersonalized medicineICU mortalityclinical decision supportbiomarker stability
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

DeepEN asks whether a machine-learning policy can personalize calorie, protein, and fluid targets for tube-fed ICU patients every four hours, and it claims the answer is yes. Trained on more than 11,000 patients from a public intensive-care database, the policy is estimated to achieve 18.8% mortality versus 22.5% for observed clinician behavior, and it also outperforms an expert guideline policy in expected return. The paper argues that this gain comes from conditioning nutrition decisions on evolving organ function, labs, vital signs, and trends rather than static per-kilogram rules, and from balancing long-term survival with short-term metabolic stability. If true, this would mean data-driven personalization could complement—not replace—guideline-based nutrition care and address part of the persistent gap between guidelines and actual practice.

Core claim

DeepEN's central claim is that a conservative offline reinforcement learning policy—specifically a dueling double deep Q-network with conservative Q-learning regularization—can learn safe, individualized enteral nutrition dosing from retrospective data. The policy selects 4-hourly combinations of weight-adjusted calories, protein, and water from a discretized space of 51 observed dose combinations, using a 102-variable state representation that includes demographics, diagnoses, vitals, labs, feeding history, and concurrent interventions like vasopressors and insulin. The paper reports the lowest calibrated mortality (18.8% ± 1.0%) among all tested policies—3.7 percentage points below the cli

What carries the argument

The argument is carried by a composite reward function paired with a conservative offline Q-learning update. The reward gives a terminal signal of +15 for survival and −15 for death, plus intermediate shaping rewards: a physiological term penalizing SOFA stagnation and worsening lactate, and a biomarker term rewarding maintenance of glucose within 140–180 mg/dL and phosphate within 2.5–4.5 mg/dL (with a bonus for improvement toward target). This reward is optimized by a dueling double DQN with conservative Q-learning regularization, which suppresses overestimation of rarely seen or unseen actions so the learned policy stays close to observed clinical practice. The evaluation then converts ex

Load-bearing premise

The headline mortality comparison assumes that the procedure translating a policy's expected score into a mortality rate produces a true counterfactual estimate, even though the score itself is built from the same survival/death signal the policy was trained on.

What would settle it

Compute the same score-to-mortality translation for a version of the policy trained without the terminal survival/death reward, keeping only the biomarker and physiology shaping rewards; if the estimated mortality advantage over clinicians disappears or shrinks significantly, the headline result would reflect the evaluation loop rather than the dosing decisions.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If DeepEN is right, ICU nutrition could move from fixed weight-based targets to dynamic, four-hourly recommendations conditioned on the patient's organ function, labs, and recent feeding response.
  • Because the learned policy outperforms behavior cloning, the benefit does not come solely from imitating good clinician decisions; the reinforcement learning optimization itself adds value.
  • The inclusion of biomarker stability in the reward means following the policy should also yield tighter glucose and phosphate control, not only a mortality difference.
  • The U-shaped deviation curves imply that both underfeeding and overfeeding relative to the learned targets are harmful, supporting a middle-range rather than a maximal-or-minimal dosing philosophy for enteral nutrition.
  • A reduction of roughly 37 deaths per 1,000 patients is comparable to or larger than the modest effect sizes seen in recent nutrition randomized trials, making personalization a potentially meaningful clinical lever if confirmed externally.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A natural next test, beyond the paper's internal evaluation, would be to run the same return-to-mortality mapping on a version of the policy trained without the terminal survival/death reward; that would reveal how much of the headline mortality advantage depends on the reward's own mortality signal.
  • The paper's action space covers only periods where enteral nutrition is actively given, so the policy is conditional on a decision to feed; a fuller decision-support tool would also need to decide when to withhold or interrupt feeding, a direction the paper itself flags.
  • The same reward-design logic—terminal survival balanced with biomarker shaping—could plausibly transfer to other slow-effect ICU interventions such as sedation titration or glycemic control, where immediate physiological targets and long-term outcomes both matter.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. DeepEN formulates enteral nutrition dosing in the ICU as an offline RL problem, using a D3QN with CQL on 11,378 MIMIC-IV patients. The state includes 102 variables, actions are four-level discretized calorie/protein/water targets, and the reward combines a terminal survival/death reward (R_term = +15/-15) with intermediate SOFA, lactate, glucose, and phosphate rewards. The paper reports that DeepEN achieves the lowest estimated mortality (18.8% vs. 22.5% for clinicians) and the highest CWPDIS return, and claims that deviations from the policy are independently associated with mortality and biomarker instability. The central evaluation, however, rests on a mortality estimator that is circular: it converts expected returns, which already include the terminal survival/death reward, into a mortality estimate via binning. This undermines the headline result.

Significance. If the central claim were valid, DeepEN would be a notable first application of offline RL to enteral nutrition, and the inclusion of a physiologically motivated reward and conservative Q-learning would be a useful contribution. The paper also makes code and data availability explicit, uses a substantial MIMIC-IV cohort, and compares against four baselines. However, the headline mortality reduction is not established: the estimated mortality in Table 2 is a model-derived calibration of the same return signal that the policy was trained to optimize, so the 18.8% vs. 22.5% result can be produced by construction even if the policy has no true mortality benefit. The 'independent association' claim in the abstract is also not supported by the unadjusted, binned analyses in Section 3.2.4. As presented, the paper is a proof-of-concept of an RL framework, but its main clinical performance claim requires a non-circular evaluation before it can be accepted.

major comments (3)
  1. [Sec. 3.2.4(3), Sec. 3.1.3, Table 2] The estimated-mortality procedure is circular. The terminal reward in Eq. (3.1.3) is +15 for ICU survival and -15 for death, so the expected return of a trajectory is a direct function of the outcome used to define mortality. Section 3.2.4(3) then bins expected returns and reads observed mortality in those bins to estimate a policy's mortality rate. A policy trained to maximize return will automatically be assigned a lower 'estimated mortality' even if it does not reduce mortality. Moreover, Table 2 compares the clinician policy's observed mortality (22.5%) with DeepEN's calibrated estimate (18.8%), which is an apples-to-oranges comparison. No validation of the bin-mapping is provided (for example, applying the same mapping to the clinician policy and checking that it reproduces 22.5%). Because this estimator is the sole basis for the headline 3.7 percentage-point reduction, the central
  2. [Sec. 3.2.4(4), Fig. 3, Abstract] The abstract states that deviation from the DeepEN policy was 'independently associated' with increased mortality and biomarker instability, but the main text presents only unadjusted, binned plots of dosage difference versus mortality (Figure 3) and versus biomarker deviation (Figure 4). There is no multivariable analysis, no adjustment for confounders such as severity of illness or concurrent treatments, and no formal statistical test for the U-shaped relationship. These plots can reflect confounding by indication (sicker patients receiving different dosages) rather than a causal effect of deviation. The word 'independently' is therefore unsupported by the evidence in the manuscript.
  3. [Table 2 and Sec. 3.2.4] The quantitative OPE results lack sufficient detail to be reproduced or interpreted. The definition of the expected-return bins is not given (e.g., number of bins, boundaries, whether bins are fit on the training or test set), and the reported confidence intervals (e.g., 3.7±0.17) do not state the resampling or uncertainty procedure. For CWPDIS, the behavior policy is a BC model fit to the same data, and policies like the expert-guidelines policy or random policy may fall outside the support of the behavior policy, making importance weights unstable. The absence of these details is particularly important because the entire performance comparison rests on these estimates.
minor comments (5)
  1. [Sec. 4.1, Table 2] There is a duplicated sentence: 'DeepEN also resulted the highest CWPDIS score among the policies, indicating superior expected returns under the learned value function (Table 2).' appears twice in the same paragraph.
  2. [Abstract vs. Sec. 4.1] The abstract reports a 4.0 percentage-point reduction (22.8% to 18.8%), while the main text and Table 2 report 22.5% and a 3.7-point reduction. Please harmonize these numbers.
  3. [Sec. 3.1.3, Eq. (1)] The intermediate reward notation is inconsistent: the definition of R_t writes R_im(t), but the text and equations use R_phys and R_bio without defining R_im. Please clarify the relationship between these terms.
  4. [Sec. 3.2.4(2), Fig. 2] The p-value for the return-mortality correlation is reported as p<0.001 without specifying the test or the correlation coefficient. Please provide the test name, the effect size, and the confidence interval.
  5. [Sec. 3.2.3, Sec. 5] The paper acknowledges the handcrafted nature of the reward and the single-center data. These limitations are appropriately stated, but they should be connected to the evaluation: a handcrafted reward that includes terminal survival/death makes the return-based mortality calibration particularly suspect, not merely a future direction.

Circularity Check

2 steps flagged

DeepEN's headline mortality reduction is an artifact of mapping returns—which already contain the terminal +15/−15 survival/death reward—back onto mortality.

specific steps
  1. fitted input called prediction [Section 3.1.3 (Rewards, Rterm(m)) and Section 3.2.4(3) (Estimated Mortality Rate); reported in Section 4.1 Table 2]
    "Rterm(m)=+rT, if m=0; −rT, if m=1 ... we set rT=15. ... First, we categorized expected returns of patient trajectories into distinct bins and used mortality occurrences within each bin to calculate the average mortality rate. The resulting relationship between expected returns and mortality is then used to estimate mortality rates for all policies based on their expected returns. ... DeepEN achieved the lowest estimated mortality among all tested policies, with a 3.7±0.17 percent lower mortality rate compared to the clinician policy (18.8% vs. 22.5%)"

    The quantity being binned (expected return) is defined using the terminal reward Rterm(m)=+15 for survival and −15 for death, so return already encodes the outcome that the procedure claims to estimate. Fitting observed mortality as a function of return and then reading off a policy's mortality from its return is a calibration of the training objective, not an independent counterfactual estimate. Any policy with higher expected return is mechanically assigned lower estimated mortality. The comparison is also asymmetric: 22.5% is the observed cohort mortality while 18.8% is a model-derived bin estimate, with no held-out validation of the bin mapping.

  2. self definitional [Section 4.2 (Return-Mortality Correlation, Figure 2)]
    "Figure 2 displays a strong negative correlation between expected returns and mortality (p <0.001). This indicates that the policy learns to associate actions leading to lower returns with higher mortality and vice versa."

    Because Rterm is +15 for survival and −15 for death, the negative return-mortality correlation is present by construction in the reward definition before any policy is learned. Presenting it as evidence that the policy 'learns' to associate low returns with mortality is a restatement of the reward function, not an independent finding about clinical alignment.

full rationale

The paper's headline claim—DeepEN reduces estimated mortality from 22.5% to 18.8%—is produced by a chain that is circular at its core. Section 3.1.3 defines the terminal reward as +15 for ICU survival and −15 for death; Section 3.2.4(3) bins expected returns and uses observed mortality within each bin to construct a return-to-mortality mapping; Section 4.1 then reads DeepEN's mortality off that mapping. Since the return used in the binning already contains the terminal survival/death reward, higher expected return and lower estimated mortality are linked by construction rather than by demonstrated counterfactual efficacy. The Figure 2 'return-mortality correlation' is similarly definitional. The comparison in Table 2 is also apples-to-oranges: the clinician baseline is the observed cohort mortality (22.5%) while DeepEN's 18.8% is a calibrated model output, and no validation of the calibration on held-out clinician trajectories is reported. Other evaluation components (biomarker target-range achievement, action distributions, CWPDIS ordering) retain some independent content, and the reward-design self-citation [12] is supportive rather than load-bearing, which is why the score is 8 rather than 10. But the central mortality-reduction claim is not established as an independent prediction.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 0 invented entities

The central claim depends on a handcrafted reward function with at least five manually set scalar parameters, on empirical quantiles for action discretization, and on an OPE procedure that re-uses the terminal mortality reward to estimate mortality. No new physical entities are introduced. The free parameters are not fitted to a held-out clinical outcome; they are designer choices embedded in the training signal.

free parameters (5)
  • Terminal reward magnitude rT = 15
    Chosen to 'preserve the dominance of the terminal outcome' (Section 3.1.3). Directly shapes the return that is later used to estimate mortality.
  • Physiological reward weights c0, c1, c2 = 0.025, 0.125, 2.0
    Fixed scalar weights 'chosen based on prior validation' (Eq. 1, Section 3.1.3). No derivation or data-fitting procedure is given.
  • Biomarker weight lambda_g, lambda_p and bonus epsilon = 1, 1, 0.2
    Hand-set to balance glucose/phosphate rewards and constrain reward magnitude (Section 3.1.3).
  • CQL alpha and discount gamma = alpha=0.5, gamma=0.99
    Selected by grid search over offline evaluation performance (Section 3.2.3), and final performance is then reported on the same evaluation metrics, creating selection bias risk.
  • Action discretization thresholds (Table 5) = Empirical quantiles (e.g., calories 0-1.91, 1.91-3.05, ... kcal/kg)
    Calorie, protein, and water bins are defined by empirical quantiles of the dataset, so the policy and all baselines operate relative to this data-driven discretization.
axioms (6)
  • domain assumption MIMIC-IV accurately records enteral nutrition administration, intervened variables, and outcomes.
    All training and evaluation rely on the completeness and accuracy of MIMIC-IV input events, labs, and mortality outcomes.
  • domain assumption The MDP constructed from 4-hourly aggregated, linearly interpolated EHR data satisfies the Markov property.
    The policy assumes that 102 state variables (including rate-of-change terms) are a sufficient statistic for future outcomes. Section 3.1.1.
  • domain assumption The off-policy value estimator CWPDIS is unbiased under the trained behavior-cloning policy.
    CWPDIS requires the behavior policy to be known/supported; the paper uses a trained BC model as behavior policy, which introduces approximation error and possible off-support bias. Section 3.2.4.
  • ad hoc to paper The return-to-mortality binning curve is a valid counterfactual mortality estimator.
    Section 3.2.4(3) assumes that observed mortality within return bins can be used to convert expected returns into estimated mortality rates for any policy. This is the load-bearing circular step.
  • ad hoc to paper The handcrafted reward function correctly encodes clinical priorities.
    Weights and target ranges (glucose [140,180] mg/dL, phosphate [2.5,4.5] mg/dL) are chosen by clinical judgment, not learned or validated. Section 3.1.3.
  • domain assumption Excluding periods without EN and restricting to the first 10 ICU days does not bias policy evaluation.
    The cohort only includes EN-active periods (Section 3.2.1), so the policy is not evaluated on feeding initiation, interruptions, or parenteral nutrition.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of DeepEN: A Deep Reinforcement Learning Framework for Personalized Enteral Nutrition in Critical Care." pith.science (2026). https://pith.science/paper/EWPVYKQQ

@misc{pith2026251008350,
  author       = {Pith},
  title        = {Pith review of: DeepEN: A Deep Reinforcement Learning Framework for Personalized Enteral Nutrition in Critical Care},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EWPVYKQQ}},
  note         = {Machine review of arXiv:2510.08350}
}
Share X Bluesky LinkedIn Reddit HN
abstract

Objective: Enteral nutrition (EN) delivery in the ICU remains suboptimal due to limited personalization and uncertainty regarding appropriate calorie, protein, and fluid targets under dynamic metabolic demands. We introduce DeepEN, a reinforcement learning (RL) framework for personalized EN optimization using electronic health record data. Methods: DeepEN was trained on over 11,000 ICU patients from MIMIC-IV to generate 4-hourly, patient-specific caloric, protein, and fluid targets. The state representation incorporated demographics, comorbidities, vital signs, laboratory values, and recent interventions. A physiologically aligned reward framework balanced biomarker stability with long-term survival. Policy learning employed a dueling double deep Q-network with Conservative Q-Learning regularization to enable safe offline training. Results: DeepEN achieved the highest estimated policy value ($V^\pi = 9.48$) and the lowest calibrated mortality (18.8 +/- 1.0%), representing a 4.0 percentage-point absolute reduction compared with clinician practice (22.8%). The policy also demonstrated superior metabolic stability, achieving the highest proportion of glucose, phosphate, and sodium values within target range. Furthermore, deviation from the DeepEN policy was independently associated with increased mortality and biomarker instability, whereas deviation from a random policy showed no such association. Interpretability analyses further indicated that recommendations were conditioned on physiologically relevant markers of organ function and metabolic status rather than static dosing heuristics. Conclusion: DeepEN demonstrates the feasibility of conservative offline RL for safe, individualized EN optimization, highlighting the potential of data-driven personalization to complement guideline-based approaches in critical care.

Figures

Figures reproduced from arXiv: 2510.08350 by Daniel Jason Tan, Dilruk Perera, Jiayang Chen, Kay Choong See, Mengling Feng.

Figure 1
Figure 1. Figure 1: Overview of the DeepEN solution pipeline. [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Mortality vs. expected returns. The shaded area represents the confidence interval. [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Dosage differences (x-axis) versus mortality (y-axis) for all policies. The shaded [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Dosage differences (x-axis) versus biomarker deviation (y-axis) for DeepEN. (A) Glu [PITH_FULL_IMAGE:figures/full_fig_p019_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Action Distributions of a) Calories, b) Protein, and c) Water for Clinician, EG, and [PITH_FULL_IMAGE:figures/full_fig_p031_5.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

44 extracted references · 3 linked inside Pith

  1. [1]

    Adeyinka, A

    A. Adeyinka, A. S. Rouster, M. Valentine, Enteric feedings, Journal Name (2018)

    2018

  2. [2]

    Preiser, Y

    J.-C. Preiser, Y. M. Arabi, M. M. Berger, M. Casaer, S. McClave, J. C. Montejo-González, S. Peake, A. Reintam Blaser, G. Van den Berghe, A. van Zanten, et al., A guide to enteral nutrition in intensive care units: 10 expert tips for the daily practice, Critical Care 25 (1) (2021) 424

    2021

  3. [3]

    Compher, A

    C. Compher, A. L. Bingham, M. McCall, J. Patel, T. W. Rice, C. Braun- schweig, L. McKeever, Guidelines for the provision of nutrition support therapy in the adult critically ill patient: The american society for par- enteral and enteral nutrition, Journal of Parenteral and Enteral Nutrition 46 (1) (2022) 12–41. 22

    2022

  4. [4]

    J. L. Bels, S. Thiessen, R. J. van Gassel, A. Beishuizen, A. D. B. Dekker, V. Fraipont, S. Lamote, D. Ledoux, C. Scheeren, E. De Waele, et al., Effect of high versus standard protein provision on functional recovery in peo- ple with critical illness (precise): an investigator-initiated, double-blinded, multicentre, parallel-group, randomised controlled tr...

    2024

  5. [5]

    D. K. Heyland, J. Patel, C. Compher, T. W. Rice, D. E. Bear, Z.-Y. Lee, V. C. González, K. O’Reilly, R. Regala, C. Wedemire, et al., The effect of higher protein dosing in critically ill patients with high nutritional risk (effort protein): an international, multicentre, pragmatic, registry-based randomised trial, The Lancet 401 (10376) (2023) 568–576

    2023

  6. [6]

    Mirhosiny, M

    M. Mirhosiny, M. Arab, P. M. Shahrbabaki, How do physicians and nurses differintheirperceivedbarrierstoeffectiveenteralnutritionintheintensive care unit?, Acute and critical care 36 (4) (2021) 342–350

    2021

  7. [7]

    Friesecke, A

    S. Friesecke, A. Schwabe, S.-S. Stecher, P. Abel, Improvement of enteral nutrition in intensive care unit patients by a nurse-driven feeding protocol, Nursing in Critical Care 19 (4) (2014) 204–210

    2014

  8. [8]

    R. J. Jarden, L. J. Sutton, A practice change initiative to improve the provision of enteral nutrition to intensive care patients, Nursing in critical care 20 (5) (2015) 242–255

    2015

  9. [9]

    C. S. Ellis, Improving nutrition in mechanically ventilated patients, Journal of Neuroscience Nursing 47 (5) (2015) 263–270

    2015

  10. [10]

    Ramaswamy, M

    T. Ramaswamy, M. P. DeWane, H. S. Dashti, M. Lau, P. E. Wischmeyer, A. Nagrebetsky, J. Sparling, Nine myths about enteral feeding in critically ill adults: an expert perspective, Advances in Nutrition (2024) 100345

    2024

  11. [11]

    Raghu, M

    A. Raghu, M. Komorowski, I. Ahmed, L. Celi, P. Szolovits, M. Ghas- semi, Deep reinforcement learning for sepsis treatment, arXiv preprint arXiv:1711.09602 (2017). 23

    Pith/arXiv arXiv 2017

  12. [12]

    D. J. Tan, Q. Xu, K. C. See, D. Perera, M. Feng, Advancing multi-organ disease care: A hierarchical multi-agent reinforcement learning framework, arXiv preprint arXiv:2409.04224 (2024)

    Pith/arXiv arXiv 2024

  13. [13]

    Peine, A

    A. Peine, A. Hallawa, J. Bickenbach, G. Dartmann, L. B. Fazlic, A. Schmeink, G. Ascheid, C. Thiemermann, A. Schuppert, R. Kindle, et al., Development and validation of a reinforcement learning algorithm to dynamically optimize mechanical ventilation in critical care, NPJ digital medicine 4 (1) (2021) 32

    2021

  14. [14]

    C. J. Watkins, P. Dayan, Q-learning, Machine learning 8 (1992) 279–292

    1992

  15. [15]

    C. Yu, J. Liu, S. Nemati, G. Yin, Reinforcement learning in healthcare: A survey, ACM Computing Surveys (CSUR) 55 (1) (2021) 1–36

    2021

  16. [16]

    V. Mnih, K. Kavukcuoglu, D. Silver, A. A. Rusu, J. Veness, M. G. Bellemare, A. Graves, M. Riedmiller, A. K. Fidjeland, G. Ostrovski, et al., Human-level control through deep reinforcement learning, nature 518 (7540) (2015) 529–533

    2015

  17. [17]

    Levine, A

    S. Levine, A. Kumar, G. Tucker, J. Fu, Offline reinforcement learn- ing: Tutorial, review, and perspectives on open problems, arXiv preprint arXiv:2005.01643 (2020)

    Pith/arXiv arXiv 2005

  18. [18]

    Van Hasselt, A

    H. Van Hasselt, A. Guez, D. Silver, Deep reinforcement learning with dou- ble q-learning, in: Proceedings of the AAAI conference on artificial intelli- gence, Vol. 30, 2016

    2016

  19. [19]

    Z. Wang, T. Schaul, M. Hessel, H. Hasselt, M. Lanctot, N. Freitas, Dueling network architectures for deep reinforcement learning, in: International conference on machine learning, PMLR, 2016, pp. 1995–2003

    2016

  20. [20]

    Hessel, J

    M. Hessel, J. Modayil, H. Van Hasselt, T. Schaul, G. Ostrovski, W. Dabney, D.Horgan, B.Piot, M.Azar, D.Silver, Rainbow: Combiningimprovements in deep reinforcement learning, in: Proceedings of the AAAI conference on artificial intelligence, Vol. 32, 2018. 24

    2018

  21. [21]

    Kumar, A

    A. Kumar, A. Zhou, G. Tucker, S. Levine, Conservative q-learning for offline reinforcement learning, Advances in neural information processing systems 33 (2020) 1179–1191

    2020

  22. [22]

    Besculides, M

    M. Besculides, M. Mazumdar, S. Phlegar, R. Freeman, S. Wilson, H. Joshi, A. Kia, K. Gorbenko, et al., Implementing a machine learning screening tool for malnutrition: insights from qualitative research applicable to other machine learning–based clinical decision support systems, JMIR Formative Research 7 (1) (2023) e42262

    2023

  23. [23]

    Raphaeli, L

    O. Raphaeli, L. Statlender, C. Hajaj, I. Bendavid, A. Goldstein, E. Robin- son, P. Singer, Using machine-learning to assess the prognostic value of early enteral feeding intolerance in critically ill patients: a retrospective study, Nutrients 15 (12) (2023) 2705

    2023

  24. [24]

    K. Hu, X. lei Deng, L. Han, S. Xiang, B. Xiong, L. Pinhu, et al., De- velopment and validation of a predictive model for feeding intolerance in intensive care unit patients with sepsis, Saudi Journal of Gastroenterology 28 (1) (2022) 32–38

    2022

  25. [25]

    Q. Chen, Y. Chen, H. Wang, J. Huang, X. Ou, J. Hu, X. Yao, L. Guan, Development and validation of a predictive model for diarrhea in icu pa- tients with enteral nutrition, Journal of Parenteral and Enteral Nutrition 47 (4) (2023) 563–571

    2023

  26. [26]

    A. R. Gosmanov, G. E. Umpierrez, Management of hyperglycemia during enteral and parenteral nutrition therapy, Current diabetes reports 13 (1) (2013) 155–162

    2013

  27. [27]

    Uribarri, Phosphorus homeostasis in normal health and in chronic kidney disease patients with special emphasis on dietary phosphorus intake., in: Seminars in dialysis, Vol

    J. Uribarri, Phosphorus homeostasis in normal health and in chronic kidney disease patients with special emphasis on dietary phosphorus intake., in: Seminars in dialysis, Vol. 20, 2007

    2007

  28. [28]

    Kalantar-Zadeh, L

    K. Kalantar-Zadeh, L. Gutekunst, R. Mehrotra, C. P. Kovesdy, R. Bross, C. S. Shinaberger, N. Noori, R. Hirschberg, D. Benner, A. R. Nissenson, 25 et al., Understanding sources of dietary phosphorus in the treatment of pa- tients with chronic kidney disease, Clinical Journal of the American Society of Nephrology 5 (3) (2010) 519–530

    2010

  29. [29]

    T. A. Ikizler, N. J. Cano, H. Franch, D. Fouque, J. Himmelfarb, K. Kalantar-Zadeh, M. K. Kuhlmann, P. Stenvinkel, P. TerWee, D. Teta, et al., Prevention and treatment of protein energy wasting in chronic kidney disease patients: a consensus statement by the international society of renal nutrition and metabolism, Kidney international 84 (6) (2013) 1096–1107

    2013

  30. [30]

    H. M. Mehanna, J. Moledina, J. Travis, Refeeding syndrome: what it is, and how to prevent and treat it, Bmj 336 (7659) (2008) 1495–1498

    2008

  31. [31]

    Nguyen, K

    N. Nguyen, K. Ching, R. Fraser, M. Chapman, R. Holloway, The rela- tionship between blood glucose control and intolerance to enteral feeding during critical illness, Intensive care medicine 33 (12) (2007) 2085–2092

    2007

  32. [32]

    diabetes care in the hospital: standards of care in diabetes—2024, Di- abetes Care 47 (Supplement_1) (2024) S295–S306

    16. diabetes care in the hospital: standards of care in diabetes—2024, Di- abetes Care 47 (Supplement_1) (2024) S295–S306

    2024

  33. [33]

    D. A. Geerse, A. J. Bindels, M. A. Kuiper, A. N. Roos, P. E. Spronk, M. J. Schultz, Treatment of hypophosphatemia in the intensive care unit: a review, Critical Care 14 (4) (2010) R147

    2010

  34. [34]

    A. E. Johnson, L. Bulgarelli, L. Shen, A. Gayles, A. Shammout, S. Horng, T. J. Pollard, S. Hao, B. Moody, B. Gow, et al., Mimic-iv, a freely accessible electronic health record dataset, Scientific data 10 (1) (2023) 1

    2023

  35. [35]

    T. J. Iwashyna, C. L. Hodgson, D. Pilcher, N. Orford, J. D. Santamarial, M. Bailey, R. Bellomo, Towards defining persistent critical illness and other varieties of chronic critical illness, Critical Care and Resuscitation 17 (3) (2015) 215–218

    2015

  36. [36]

    P. E. Wischmeyer, D. E. Bear, M. M. Berger, E. De Waele, J. Gunst, S. A. McClave, C. M. Prado, Z. Puthucheary, E. J. Ridley, G. Van den 26 Berghe, et al., Personalized nutrition therapy in critical care: 10 expert recommendations, Critical Care 27 (1) (2023) 261

    2023

  37. [37]

    Stoian, A

    M. Stoian, A. Andone, S. R. Bândilă, D. Onis,or, D.-F. Babă, R. Niculescu, A. Stoian, L. Azamfirei, Personalized nutrition strategies for patients in the intensive care unit: A narrative review on the future of critical care nutrition, Nutrients 17 (10) (2025) 1659

    2025

  38. [38]

    Thomas, E

    P. Thomas, E. Brunskill, Data-efficient off-policy policy evaluation for re- inforcement learning, in: International conference on machine learning, PMLR, 2016, pp. 2139–2148

    2016

  39. [39]

    X. Wu, R. Li, Z. He, T. Yu, C. Cheng, A value-based deep reinforcement learning model with human expertise in optimal treatment of sepsis, NPJ Digital Medicine 6 (1) (2023) 15

    2023

  40. [40]

    Komorowski, L

    M. Komorowski, L. A. Celi, O. Badawi, A. C. Gordon, A. A. Faisal, The artificial intelligence clinician learns optimal treatment strategies for sepsis in intensive care, Nature medicine 24 (11) (2018) 1716–1720

    2018

  41. [41]

    D. E. Bear, L. Wandrag, J. L. Merriweather, B. Connolly, N. Hart, M. P. Grocott, E.R.A.C.I.P.G.E.investigators, Theroleofnutritionalsupport in the physical and functional recovery of critically ill patients: a narrative review, Critical Care 21 (1) (2017) 226. 27 Appendix

    2017

  42. [42]

    Cohort Information Table 3: Cohort details Cohort % Female Mean Age (years) Mean ICU Stay (hours) Total Population (n) Overall 41.63 64.92 241 11378Non-Survivors 43.19 68.76 258 2556Survivors 41.18 63.81 237 8822

  43. [43]

    Expert Guidelines Policy Definition These nutritional targets are based on the 2021 ASPEN requirements for enteral nutrition [3]. Original protein targets have been altered slightly with more recent evidence-based literature recommending a low-dosing period during the early-acute phase of critical illness [36, 37]. 3.1. Calories •For patients with BMI<30:...

    2021

  44. [44]

    Comparison of Action Distributions Figure 5: Action Distributions of a) Calories, b) Protein, and c) Water for Clinician, EG, and DeepEN policies. 31

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.