Pith. sign in

REVIEW 3 major objections 5 minor 56 references

Identification of Pediatric Sepsis Subphenotypes for Enhanced Machine Learning Predictive Performance: A Latent Profile Analysis

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Subtyping pediatric sepsis lifts prediction AUC from .918 to .998.

desk verdict The pediatric sepsis LPA subphenotypes may be interesting, but the ML performance claim is confounded by label leakage and an asymmetric baseline. read the letter →

arxiv 1908.09038 v1 pith:EDSWBREI submitted 2019-08-23 stat.ML cs.LGq-bio.QMstat.AP

Tom Velez , Tony Wang , Ioannis Koutroulis , James Chamberlain , Amit Uppal , Seife Yohannes , Tim Tschampel , Emilia Apostolova This is my paper · ORCID
classification stat.MLcs.LGq-bio.QMstat.AP
keywords pediatricsepsismortalityrisklatentprofileanalysismachinelearningsubphenotypeselectronichealthrecordsgradientboostingrandomforest
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Sepsis in children is not one condition for machine-learning purposes, this paper argues. Latent profile analysis applied to 134 pediatric sepsis patients from a 6,446-patient cohort separates them into four subphenotypes from routine vitals and labs, and for two of those groups—the respiratory-dysfunction and neurological-dysfunction profiles—models trained on the partition score significantly higher on AUC (a standard accuracy score, 1 being perfect) than one model trained on all sepsis patients together. The headline comparison is the neurological profile, whose mortality is 22.2 percent: with the first 24 hours of data the profile-specific model reaches AUC 0.998 versus 0.918 for the all-phenotype model. The authors take this as evidence that reducing training-data heterogeneity by subtyping can make early recognition of the highest-risk septic children more reliable.

What carries the argument

Latent profile analysis (LPA) is the load-bearing mechanism: a Gaussian finite mixture model, fitted by expectation-maximization on Yeo-Johnson-transformed clinical variables, with the number of profiles selected by the Bayesian information criterion. The chosen four-profile VEI model has diagonal covariance, varying volume, and equal shape. LPA's job is to turn the heterogeneous sepsis population into homogeneous training partitions before any supervised learning; gradient-boosted machines and random forests, SMOTE resampling, and median imputation are then held fixed so that the only intended difference between the all-phenotype and profile-specific pipelines is the homogeneity of the training data. The near-perfect profile-4 result is carried mainly by the Glasgow Coma Scale and FiO2, which rank as the top two features in that model.

What would settle it

Run the same experiment with profiles fitted only on the training half, and evaluate both models on the same held-out profile-4 cases; if the profile-specific AUC drops toward the all-phenotype value of 0.918, the reported 0.998 is an artifact of leakage or label-feature overlap rather than a heterogeneity-reduction benefit. Excluding GCS from the profile definition would be a second decisive check.

Watch

Extended reading notes

Core claim

The paper's central claim is that latent profile analysis identifies four clinically meaningful pediatric sepsis subphenotypes and that targeting models to those subphenotypes improves predictive performance. Profile 4, defined by neurological dysfunction and the lowest Glasgow Coma Scale scores, had the highest mortality (22.2%) and was predicted with AUC 0.998 using 24-hour data, against AUC 0.918 for the all-phenotype model (p < 0.0001); profile 2, defined by respiratory dysfunction and low PaO2, was predicted with AUC 0.976 (p = 0.0077) on the same 24-hour window, and AUC 0.994 on whole-stay data (p = 0.0029). The authors report that these gains appear only once enough data accumulate—profile 4's 6-hour gain is not significant (p = 0.0915)—and that feature importance shifts across profiles, with GCS and FiO2 leading for profile 4 and age and creatinine leading overall.

Load-bearing premise

The result stands on the assumption that the higher profile-specific accuracy comes from training on a more homogeneous group, which requires the profile-specific and all-phenotype models to be tested on the same children and the profiles to be defined before the test set is examined—conditions the paper does not explicitly satisfy.

Editorial extensions

If this is right

  • If the reported gains are real, the same LPA-before-training workflow should transfer to other heterogeneous critical-care syndromes, since the argument is about reducing training-data heterogeneity rather than about a sepsis-specific biomarker.
  • Profile 4, the neurological-dysfunction group with 22.2% mortality, becomes separable from non-septic controls with near-perfect AUC using routine vitals and labs within 24 hours, so it is a plausible target for a high-risk early-alert system.
  • The benefit is timing-dependent: profile 4's improvement is not statistically significant with 6-hour data but is highly significant with 12- and 24-hour data, meaning the practical value depends on how quickly the required measurements are available.
  • The shift in feature importance—from age and creatinine in the all-phenotype model to GCS and FiO2 in profile 4—supports the paper's interpretation that the subphenotypes are physiologically distinct, not just different severity strata.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural next step the paper does not take is a two-stage clinical pipeline: first assign a patient to a profile from early data, then run the profile-specific sepsis model; the end-to-end accuracy of that pipeline remains untested.
  • Because profile 4's label is defined by the lowest Glasgow Coma Scale and GCS is also the model's top feature, some of the AUC gap may reflect the label being nearly a restatement of the predictor; removing GCS from the profile definition would isolate what the LPA partition itself contributes.
  • The strongest external validation would be to apply the identical protocol to adult sepsis data, where phenotype differences in treatment response have already been reported, and to report PPV and calibration alongside AUC, since PPV is the clinically limiting metric in this study (0.421 for profile 4 at 24 hours).
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The manuscript applies latent profile analysis (LPA) to 134 pediatric sepsis patients drawn from a 6,446-patient retrospective cohort, identifies four sepsis subphenotypes, and then builds gradient-boosted machine and random forest models for each subphenotype separately, with all nonsepsis patients as controls. The central claim is that subphenotype-targeted models achieve significantly higher AUC than a model trained on all sepsis patients, e.g., AUC = 0.998 with 24-hour data for profile 4 versus AUC = 0.918 for the all-phenotype model (p < 0.0001), and that this demonstrates that LPA-based partitioning improves predictive performance by reducing training-data heterogeneity. The paper reports that profiles 2 and 4, characterized by respiratory and neurological dysfunction respectively, show the largest improvements and have higher mortality and vasopressor use.

Significance. If the central comparison were valid, the paper would offer a clinically relevant demonstration that unsupervised subphenotyping can improve machine-learning sepsis screening in pediatrics, building on prior adult sepsis subphenotype work. The study has useful strengths: it uses routinely available EHR data, reports detailed variable statistics, provides SQL screening criteria, and explicitly acknowledges limitations such as missing data and small sample size. However, the load-bearing claim rests on a comparison that is confounded in two independent ways: the profile-specific models are evaluated on a narrower, more separable target than the all-phenotype model, and the profile-4 label is defined by a variable (GCS) that also dominates the profile-4 predictive model. The reported AUC gap therefore does not isolate the benefit of reduced training-data heterogeneity, and the manuscript does not provide the reanalysis needed to support its primary conclusion. The paper also lacks external validation, and the within-cohort use of LPA before the train/test split introduces label leakage into the predictive experiments.

major comments (3)
  1. [§3.2, §3.3, Table 4] The central comparison is confounded by an asymmetry in the evaluation target. For each subphenotype, the positive class is a small subset of sepsis patients (e.g., 27 profile-4 patients) and the negative class is the full nonsepsis control group, while other sepsis profiles are excluded from the test set. The 'all phenotypes' model is evaluated on all sepsis patients against the same controls. Thus the profile-4 AUC of 0.998 measures discrimination of a GCS-defined subset against more than 6,000 controls, whereas the all-phenotype AUC of 0.918 measures discrimination of the entire heterogeneous sepsis population. These are different classification tasks, and the DeLong p-values reported in Table 4, which compare models on different test sets, are not valid evidence for a benefit of LPA. A fair comparison would score the all-phenotype model on the same profile-4-versus-nonsepsis test set as the profile-4 model, or equivalently would evaluate all models on the full sepsis-versus-control test set.
  2. [§2.2, Table 2, Appendix Table A2] The near-perfect AUC for profile 4 is substantially explained by circularity: LPA was fit on all 134 sepsis patients using post-onset clinical variables before any train/test split (§2.2), and Table 2 shows that profile 4 is defined by a markedly low minimum GCS (median 6.0 vs 14–15 in other profiles). The profile-4 predictive model uses the same feature set, and Appendix Table A2 lists GCS as the top feature with importance 100. With median imputation for missing GCS and the same cohort used for LPA and evaluation, the model can approach perfect discrimination by thresholding on the very variable that was used to assign the profile label. This is not a test of whether LPA partitions improve generalization; it is at least partly a restatement of the label-defining variable. The authors should either use temporally separated data (e.g., LPA derived from a discovery cohort and applied to an independent cohort) or explicitly exclude label-defining variables from the predictive models.
  3. [§3.2, Table 4] The reported statistical significance does not support the heterogeneity-reduction claim because the sample sizes and test compositions differ across models. Each profile-specific model has only 27–39 positive patients, and a 70/30 split leaves roughly 8–12 positive test cases per profile. The bootstrap confidence intervals for sensitivity and specificity are consequently very wide (e.g., profile 2 sensitivity 1.0 [1.0, 1.0] in the 24-hour row despite only about 10 test positives), and the DeLong comparison is applied to ROC curves estimated on different subsets of patients. The claim that partitioning reduces heterogeneity would require a matched evaluation protocol, ideally with repeated cross-validation and with the same test patients scored by both the profile-specific and all-phenotype models. As reported, the results do not establish that LPA-derived partitioning improves predictive performance over a single model on the full population.
minor comments (5)
  1. [Abstract and §1] The abstract states a cohort of 6,446 patients while the full text uses 6,446 in one place and 6446 in another; the number should be consistent throughout.
  2. [§2.1 and §3.1] The time origin for feature extraction is inconsistent: §2.1 defines sepsis onset via age-adjusted SOFA, and §2.2 says LPA uses post-onset data, but §3.1 describes features extracted over 'the first 6, 12, 24 hours following admission.' The authors should clarify whether the predictive-model features are anchored to admission or to sepsis onset, since this affects the clinical interpretation of the early-warning results.
  3. [§4] The Discussion states 'At an AUC of .98 compared to AUC of .88 (p < .0001) for Profile 4 using data within 12 hours following admission,' but Table 4 reports 12-hour AUCs of 0.989 for profile 4 and 0.885 for all phenotypes; the numbers should match the table.
  4. [Table A2] The variable-importance ranking skips ranks 13, 14, 20, and 21 for the 'all phenotypes' column without explanation; the table should either include all ranks or note omitted variables.
  5. [General] The paper does not report the exact number of patients used in each train/test split for the profile-specific models, nor the class-imbalance ratios after SMOTE. Adding this information would improve reproducibility.

Circularity Check

2 steps flagged · score 6.0 of 10

Profile-4 'prediction' restates its GCS-based definition, and LPA labels are fit on the full cohort before the train/test split.

  1. self definitional [Section 2.3 and Table 2 (profile definition); Section 3.3 and Table A2 (feature importance)]
    "Importance of features in predicting phenotypes were found to be different across phenotypes (Appendix Table A2). GCS and FIO2 ranked top in predicting phenotype 4, while age and creatinine ranked top in predicting other phenotypes."

    Subphenotype 4 is defined by having the lowest GCS total score: Table 2 lists median minimum GCS 6.0 for profile 4 versus 14–15 for the other profiles, and Section 2.3 calls profile 4 'neurological dysfunction (lowest GCS total score)'. The profile-4 predictive model then ranks GCS as its most important feature (importance 100). Thus the near-perfect AUC (0.998) is largely a threshold on the same variable used to construct the label: predicting a label from its defining feature is self-definitional, not independent evidence that LPA-based partitioning improves predictive performance.

  2. fitted input called prediction [Sections 2.2, 3.1, and 3.2]
    "All available post-onset clinical data and lab measurements of 134 sepsis patients were considered as profile-defining variables in the LPA modelling; profiling was conducted without consideration of clinical outcomes. ... For predictive modeling, training data was partitioned into 4 groups according to profile assignment. ... Data was split into a training (70%) and test set (30%)."

    LPA is fit on all 134 sepsis patients before any train/test split, and the resulting profile assignments are then used to define both training and test partitions. The test-set profile labels are therefore not out-of-sample: they are derived from a model that saw the entire cohort and the full-stay features. The profile-specific test AUC measures the model's ability to recover labels constructed from the same data, so the comparison against the all-phenotype model does not independently validate the benefit of reduced training-data heterogeneity; it partly restates the clustering input.

full rationale

The central claim—that LPA-identified subphenotype-targeted models outperform an all-phenotype model—is not circular in the sense of a theorem deriving itself, but the headline result (profile 4, AUC 0.998) is largely self-definitional. Profile 4 is characterized by the lowest GCS total score, and GCS is the top feature in the profile-4 model, so a model can approach perfect AUC by thresholding on the same variable used to define the label. In addition, LPA was fit on all 134 sepsis patients before the train/test split, so the profile labels used in the test set are not independent of the data used to create them; this is label leakage rather than a fitted parameter renamed as a prediction, but it has the same effect of forcing the profile-specific result by construction. The Table 4 comparison is also asymmetric: the all-phenotype AUC is for 'any sepsis' on the full test set, while the profile-4 AUC is for a small, easily separable subset (27 patients with very low GCS), with other sepsis profiles excluded. A fair test would evaluate the all-phenotype model on the same profile-4-positive cases and the same controls. The paper honestly acknowledges the small sample and the need for external validation, but that does not repair the circularity of the profile-4 headline. Other aspects, such as the LPA clustering itself and differences in feature importance across profiles, are not circular. Self-citations in the paper are motivational rather than load-bearing. Overall, partial circularity: score 6.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

The central analysis rests on the LPA model and its fitted profiles, on the operationalized sepsis definition, and on the assumption that the ML comparison isolates the effect of subtyping. The latter assumption is not tested and is likely violated. The ledger lists the fitted parameters and assumptions that are not independently justified.

free parameters (4)
  • Number of latent profiles = 4
    Selected by BIC across model types and component counts; not independently validated or externally replicated.
  • Mixture model parameters (component means, covariance, mixing weights) = Not fully reported; means/SDs per profile in Table 2
    Fitted by EM to the 134 sepsis patients; the central claim depends on these fitted profiles.
  • Median imputation values and SMOTE resampling settings = Not reported
    Imputation medians computed from the cohort; SMOTE parameters unspecified, affecting reproducibility and bias.
  • ML hyperparameters (trees, depth, learning rate, etc.) = Not reported
    Tuned by cross-validation but final values are not given, making the specific AUC results not reproducible.
assumptions (5)
  • domain assumption The population is a finite Gaussian mixture with diagonal covariance (VEI model)
    LPA assumes approximate normality after Yeo-Johnson transformation and a specific covariance structure; this is a strong distributional assumption for EHR vitals and labs.
  • domain assumption Sepsis-3 organ dysfunction can be identified from ICD-9/10 codes and mechanical ventilation codes
    The case definition in Section 2.1 follows Zhang and may misclassify sepsis; no chart review validation is reported.
  • domain assumption Age-adjusted SOFA onset time from EHR data marks the true sepsis onset
    Onset time is defined as the first abnormal age-adjusted SOFA component; measurement timing and data quality affect this.
  • ad hoc to paper Missing values can be handled by median imputation without biasing profile identification
    Several LPA variables have more than 40% missingness (PaO2 65.7%, pH 57.5%); the authors acknowledge this limitation but still use median imputation.
  • standard math BIC selects the correct number of latent profiles
    BIC consistency is assumed for model selection; with n=134 and 23 variables this is fragile.
invented entities (1)
  • Pediatric sepsis subphenotypes 1 to 4
    purpose: To partition patients into homogeneous groups for targeted ML prediction and potential treatment triage
    The four latent profiles are induced from the same 134-patient cohort; no external dataset, biomarker, or prospective outcome confirms their existence. They are not validated against independent data, so they are hypothesized constructs rather than established entities.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Identification of Pediatric Sepsis Subphenotypes for Enhanced Machine Learning Predictive Performance: A Latent Profile Analysis." pith.science (2026). https://pith.science/paper/EDSWBREI

@misc{pith2026190809038,
  author       = {Pith},
  title        = {Pith review of: Identification of Pediatric Sepsis Subphenotypes for Enhanced Machine Learning Predictive Performance: A Latent Profile Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EDSWBREI}},
  note         = {Machine review of arXiv:1908.09038}
}
read the original abstract

Background: While machine learning (ML) models are rapidly emerging as promising screening tools in critical care medicine, the identification of homogeneous subphenotypes within populations with heterogeneous conditions such as pediatric sepsis may facilitate attainment of high-predictive performance of these prognostic algorithms. This study is aimed to identify subphenotypes of pediatric sepsis and demonstrate the potential value of partitioned data/subtyping-based training. Methods: This was a retrospective study of clinical data extracted from medical records of 6,446 pediatric patients that were admitted at a major hospital system in the DC area. Vitals and labs associated with patients meeting the diagnostic criteria for sepsis were used to perform latent profile analysis. Modern ML algorithms were used to explore the predictive performance benefits of reduced training data heterogeneity via label profiling. Results: In total 134 (2.1%) patients met the diagnostic criteria for sepsis in this cohort and latent profile analysis identified four profiles/subphenotypes of pediatric sepsis. Profiles 1 and 3 had the lowest mortality and included pediatric patients from different age groups. Profile 2 were characterized by respiratory dysfunction; profile 4 by neurological dysfunction and highest mortality rate (22.2%). Machine learning experiments comparing the predictive performance of models derived without training data profiling against profile targeted models suggest statistically significant improved performance of prediction can be obtained. For example, area under ROC curve (AUC) obtained to predict profile 4 with 24-hour data (AUC = .998, p < .0001) compared favorably with the AUC obtained from the model considering all profiles as a single homogeneous group (AUC = .918) with 24-hour data.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 56 canonical work pages

  1. [1]

    Pediatric Sepsis - Part I: ‘Children are not small adults!,’

    D. S. Wheeler, H. R. Wong, and B. Zingarelli, “Pediatric Sepsis - Part I: ‘Children are not small adults!,’” Open Inflamm. J., vol. 4, no. 513, pp. 4–15, 2011

    work page 2011

  2. [2]

    The Host Response to Sepsis and Developmental Impact,

    J. Wynn, T. T. Cornell, H. R. Wong, T. P. Shanley, and D. S. Wheeler, “The Host Response to Sepsis and Developmental Impact,” Pediatrics, vol. 125, no. 5, pp. 1031–1041, 2010

    work page 2010

  3. [3]

    Introduction to Pediatric Sepsis.,

    D. S. Wheeler, “Introduction to Pediatric Sepsis.,” Open Inflamm. J., vol. 4, no. Suppl 1-M, pp. 1–3, 2011

    work page 2011

  4. [4]

    Management of Neonates With Suspected or Proven Early-Onset Bacterial Sepsis,

    R. A. Polin, “Management of Neonates With Suspected or Proven Early-Onset Bacterial Sepsis,” Pediatrics, vol. 129, no. 5, pp. 1006–1015, 2012

    work page 2012

  5. [5]

    Pediatric severe sepsis in U.S. children’s hospitals,

    F. Balamuth et al., “Pediatric severe sepsis in U.S. children’s hospitals,” Pediatr. Crit. Care Med., 2014

    work page 2014

  6. [6]

    Global epidemiology of pediatric severe sepsis: the sepsis prevalence, outcomes, and therapies study,

    S. L. Weiss et al., “Global epidemiology of pediatric severe sepsis: the sepsis prevalence, outcomes, and therapies study,” Am. J. Respir. Crit. Care Med., vol. 191, no. 10, pp. 1147–1157, 2015

    work page 2015

  7. [7]

    Pediatric severe sepsis: current trends and outcomes from the pediatric health information systems database*.,

    A. Ruth, C. E. McCracken, J. D. Fortenberry, M. Hall, H. K. Simon, and K. B. Hebbar, “Pediatric severe sepsis: current trends and outcomes from the pediatric health information systems database*.,” Pediatr. Crit. Care Med., vol. 15, no. 9, pp. 828–38, 2014

    work page 2014

  8. [8]

    W. P. Witt, A. J. Weiss, and A. Elixhauser, Overview of Hospital Stays for Children in the United States, 2012: Statistical Brief #187. 2006

    work page 2012

Show all 56 references
  1. [9]

    Epidemiology of pediatric hospitalizations at general hospitals and freestanding children’s hospitals in the United States,

    J. A. K. Leyenaar, S. L. Ralston, M. S. Shieh, P. S. Pekow, R. Mangione-Smith, and P. K. Lindenauer, “Epidemiology of pediatric hospitalizations at general hospitals and freestanding children’s hospitals in the United States,” J. Hosp. Med., 2016

    2016

  2. [10]

    Early recognition and management of septic shock in children,

    P. Biban, M. Gaffuri, S. Spaggiari, F. Zaglia, A. Serra, and P. Santuz, “Early recognition and management of septic shock in children,” Pediatric Reports. 2012

    2012

  3. [11]

    Adherence to PALS Sepsis Guidelines and Hospital Length of Stay,

    R. Paul, M. I. Neuman, M. C. Monuteaux, and E. Melendez, “Adherence to PALS Sepsis Guidelines and Hospital Length of Stay,” Pediatrics, 2012

    2012

  4. [12]

    Protocolized Treatment Is Associated with Decreased Organ Dysfunction in Pediatric Severe Sepsis,

    F. Balamuth et al., “Protocolized Treatment Is Associated with Decreased Organ Dysfunction in Pediatric Severe Sepsis,” Pediatr. Crit. Care Med., 2016

    2016

  5. [13]

    Resuscitation Bundle in Pediatric Shock Decreases Acute Kidney Injury and Improves Outcomes,

    A. Akcan Arikan, E. A. Williams, J. M. Graf, C. E. Kennedy, B. Patel, and A. T. Cruz, “Resuscitation Bundle in Pediatric Shock Decreases Acute Kidney Injury and Improves Outcomes,” J. Pediatr., 2015

    2015

  6. [14]

    National estimates of emergency department visits for pediatric severe sepsis in the United States,

    S. Singhal, M. W. Allen, J.-R. McAnnally, K. S. Smith, J. P. Donnelly, and H. E. Wang, “National estimates of emergency department visits for pediatric severe sepsis in the United States,” PeerJ, 2013

    2013

  7. [15]

    Clinical practice parameters for hemodynamic support of pediatric and neonatal patients in septic shock,

    J. A. Carcillo and A. I. Fields, “Clinical practice parameters for hemodynamic support of pediatric and neonatal patients in septic shock,” Crit. Care Med., 2002

    2002

  8. [16]

    Continuum of care in pediatric sepsis: a prototypical acute care delivery model,

    R. Vidrine, M. Ateya, and E. Stalets, “Continuum of care in pediatric sepsis: a prototypical acute care delivery model,” Transl. Pediatr., vol. 7, no. 4, pp. 253–261, 2018

    2018

  9. [17]

    Systemic inflammatory response in the pediatric emergency department: a common phenomenon that does not predict severe illness,

    C. Foo, G. Sangha, J. Seabrook, and J. Foster, “Systemic inflammatory response in the pediatric emergency department: a common phenomenon that does not predict severe illness,” Crit. Care, 2014

    2014

  10. [18]

    Designing a Pediatric Severe Sepsis Screening Tool,

    R. J. Sepanski, S. A. Godambe, C. D. Mangum, C. S. Bovat, A. L. Zaritsky, and S. H. Shah, “Designing a Pediatric Severe Sepsis Screening Tool,” Front. Pediatr., vol. 2, 2014

    2014

  11. [19]

    Designing a pediatric severe sepsis screening tool.,

    R. J. Sepanski, S. a Godambe, C. D. Mangum, C. S. Bovat, A. L. Zaritsky, and S. H. Shah, “Designing a pediatric severe sepsis screening tool.,” Front. Pediatr., vol. 2, no. June, p. 56, 2014

    2014

  12. [20]

    The third international consensus definitions for sepsis and septic shock (sepsis -3),

    M. Singer et al., “The third international consensus definitions for sepsis and septic shock (sepsis -3),” JAMA - Journal of the American Medical Association, vol. 315, no. 8. pp. 801–810, 2016

    2016

  13. [21]

    Prognostic accuracy of age -adapted SOFA, SIRS, PELOD-2, and qSOFA for in-hospital mortality among children with suspected infection admitted to the intensive care unit,

    L. J. Schlapbach, L. Straney, R. Bellomo, G. MacLaren, and D. Pilcher, “Prognostic accuracy of age -adapted SOFA, SIRS, PELOD-2, and qSOFA for in-hospital mortality among children with suspected infection admitted to the intensive care unit,” Intensive Care Med., 2017

    2017

  14. [22]

    Medical decision support using machine learning for early detection of late-onset neonatal sepsis,

    S. Mani et al., “Medical decision support using machine learning for early detection of late-onset neonatal sepsis,” J. Am. Med. Informatics Assoc., 2014

    2014

  15. [23]

    Pediatric Severe Sepsis Prediction Using Machine Learning,

    T. Desautels et al., “Pediatric Severe Sepsis Prediction Using Machine Learning,” bioRxiv, 2017

    2017

  16. [24]

    Raising concerns about the Sepsis-3 definitions,

    M. Sartelli et al., “Raising concerns about the Sepsis-3 definitions,” World Journal of Emergency Surgery. 2018

    2018

  17. [25]

    Predictive Learning in the Presence of Heterogeneity and Limited,

    A. Karpatne, A. Khandelwal, S. Boriah, and V. Kumar, “Predictive Learning in the Presence of Heterogeneity and Limited,” in Proceedings of the 2014 SIAM International Conference on Data Mining , 2014

    2014

  18. [26]

    The role of artificial intelligence in precision medicine,

    B. Mesko, “The role of artificial intelligence in precision medicine,” Expert Rev. Precis. Med. Drug Dev., 2017

    2017

  19. [27]

    Precision medicine for all? Challenges and opportunities for a precision medicine approach to critical illness,

    C. W. Seymour et al., “Precision medicine for all? Challenges and opportunities for a precision medicine approach to critical illness,” Critical Care. 2017

    2017

  20. [28]

    Latent Class Analysis: An Alternative Perspective on Subgroup Analysis in Prevention and Treatment,

    S. T. Lanza and B. L. Rhoades, “Latent Class Analysis: An Alternative Perspective on Subgroup Analysis in Prevention and Treatment,” Prev. Sci., 2013

    2013

  21. [29]

    A latent profile analysis of college students’ achievement goal orientation,

    D. A. Pastor, K. E. Barron, B. J. Miller, and S. L. Davis, “A latent profile analysis of college students’ achievement goal orientation,” Contemp. Educ. Psychol., 2007

    2007

  22. [30]

    An introduction to latent variable mixture modeling (Part 1): Overview and cross-sectional latent class and latent profile analyses,

    K. S. Berlin, N. A. Williams, and G. R. Parra, “An introduction to latent variable mixture modeling (Part 1): Overview and cross-sectional latent class and latent profile analyses,” Journal of Pediatric Psychology. 2014

    2014

  23. [31]

    Six subphenotypes in septic shock: Latent class analysis of the PROWESS Shock study,

    B. Gårdlund, N. O. Dmitrieva, C. F. Pieper, S. Finfer, J. C. Marshall, and B. Taylor Thompson, “Six subphenotypes in septic shock: Latent class analysis of the PROWESS Shock study,” J. Crit. Care, 2018

    2018

  24. [32]

    Identification of subclasses of sepsis that showed different clinical outcomes and responses to amount of fluid resuscitation: a latent profile analysis,

    Z. Zhang, G. Zhang, H. Goyal, and Y. Hong, “Identification of subclasses of sepsis that showed different clinical outcomes and responses to amount of fluid resuscitation: a latent profile analysis,” Crit Care, vol. 22, no. 347, pp. 1–11, 2018

    2018

  25. [33]

    Identification of three classes of acute respiratory distress syndrome using latent class analysis,

    Z. Zhang, “Identification of three classes of acute respiratory distress syndrome using latent class analysis,” PeerJ, 2018

    2018

  26. [34]

    Incorporating Prior Domain Knowledge Into Inductive Machine Learning,

    T. Yu, T. Jan, S. Simoff, and J. Debenham, “Incorporating Prior Domain Knowledge Into Inductive Machine Learning,” Mach. Learn., 2007

    2007

  27. [35]

    Integrating prior knowledge into deep learning,

    M. Diligenti, S. Roychowdhury, and M. Gori, “Integrating prior knowledge into deep learning,” in Proceedings - 16th IEEE International Conference on Machine Learning and Applications, ICMLA 2017 , 2018

    2017

  28. [36]

    Semantically Enhanced Dynamic Bayesian Network or Detecting Sepsis Mortality Risk in ICU Patients with Infection,

    J. Wang, T., Velez, T., Apostolova, E., Tschampel, T., Ngo, T., Hardison, “Semantically Enhanced Dynamic Bayesian Network or Detecting Sepsis Mortality Risk in ICU Patients with Infection,” Cornell Univ. Libr., 2017

    2017

  29. [37]

    Estimating the Dimension of a Model,

    G. Schwartz, “Estimating the Dimension of a Model,” Ann. Stat., pp. 461–464, 6AD

  30. [38]

    Model-based clustering, discriminant analysis, and density estimation,

    C. Fraley and A. E. Raftery, “Model-based clustering, discriminant analysis, and density estimation,” J. Am. Stat. Assoc., 2002

    2002

  31. [39]

    A new family of power transformations to improve normality or symmetry,

    I. N. K. Yeo and R. A. Johnson, “A new family of power transformations to improve normality or symmetry,” Biometrika, 2000

    2000

  32. [40]

    Gradient boosting machines, a tutorial,

    A. Natekin and A. Knoll, “Gradient boosting machines, a tutorial,” Front. Neurorobot., 2013

    2013

  33. [41]

    Random forests,

    A. Cutler, D. R. Cutler, and J. R. Stevens, “Random forests,” in Ensemble Machine Learning: Methods and Applications, 2012

    2012

  34. [42]

    SMOTE: Synthetic minority over -sampling technique,

    N. V. Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “SMOTE: Synthetic minority over -sampling technique,” J. Artif. Intell. Res., 2002

    2002

  35. [43]

    Bootstrap confidence intervals: when, which, what? A practical guide for medical statisticians.,

    J. Carpenter and J. Bithell, “Bootstrap confidence intervals: when, which, what? A practical guide for medical statisticians.,” Stat. Med., vol. 19, no. 9, pp. 1141–64, May 2000

    2000

  36. [44]

    Comparing the Areas under Two or More Correlated Receiver Operating Characteristic Curves: A Nonparametric Approach,

    E. R. DeLong, D. M. DeLong, and D. L. Clarke-Pearson, “Comparing the Areas under Two or More Correlated Receiver Operating Characteristic Curves: A Nonparametric Approach,” Biometrics, 1988

    1988

  37. [45]

    Fluid resuscitation in human sepsis: Time to rewrite history?,

    L. Byrne and F. Van Haren, “Fluid resuscitation in human sepsis: Time to rewrite history?,” Annals of Intensive Care. 2017

    2017

  38. [46]

    Interventions for Pediatric Sepsis and Their Impact on Outcomes: A Brief Review,

    L. Watkins, “Interventions for Pediatric Sepsis and Their Impact on Outcomes: A Brief Review,” Healthcare, 2018

    2018

  39. [47]

    Derivation, Validation, and Potential Treatment Implications of Novel Clinical Phenotypes for Sepsis,

    C. W. Seymour et al., “Derivation, Validation, and Potential Treatment Implications of Novel Clinical Phenotypes for Sepsis,” JAMA - J. Am. Med. Assoc., 2019

    2019

  40. [48]

    K-means-based consensus clustering: A unified view,

    J. Wu, H. Liu, H. Xiong, J. Cao, and J. Chen, “K-means-based consensus clustering: A unified view,” IEEE Trans. Knowl. Data Eng., 2015

    2015

  41. [49]

    A Comparison of Segment Retention Criteria for Finite Mixture Logit Models,

    R. L. Andrews and I. S. Currim, “A Comparison of Segment Retention Criteria for Finite Mixture Logit Models,” J. Mark. Res., 2003

    2003

  42. [50]

    Mixture Models: Latent Profile and Latent Class Analysis,

    D. Oberski, “Mixture Models: Latent Profile and Latent Class Analysis,” 2016

    2016

  43. [51]

    Exploiting missing clinical data in Bayesian network modeling for predicting m edical problems.,

    J. Lin and P. J. Haug, “Exploiting missing clinical data in Bayesian network modeling for predicting m edical problems.,” J. Biomed. Inform., vol. 41, no. 1, pp. 1–14, 2008

    2008

  44. [52]

    Data management by using R: big data clinical research series,

    Z. Zhang, “Data management by using R: big data clinical research series,” Ann Transl Med, 2015

    2015

  45. [53]

    Association between early lactate levels and 30-day mortality in clinically suspected sepsis in children,

    H. F. Scott, L. Brou, S. J. Deakyne, A. Kempe, D. L. Fairclough, and L. Bajaj, “Association between early lactate levels and 30-day mortality in clinically suspected sepsis in children,” JAMA Pediatr., 2017

    2017

  46. [54]

    Clinical practice parameters for hemodynamic support of pediatric and neonatal septic shock: 2007 update from the American College of Critical Care Medicine,

    J. Brierley et al., “Clinical practice parameters for hemodynamic support of pediatric and neonatal septic shock: 2007 update from the American College of Critical Care Medicine,” Crit. Care Med., 2009

    2007

  47. [55]

    Pediatric Sepsis Biomarker Risk Model-II: Redefining the Pediatric Sepsis Biomarker Risk Model with Septic Shock Phenotype,

    H. R. Wong et al., “Pediatric Sepsis Biomarker Risk Model-II: Redefining the Pediatric Sepsis Biomarker Risk Model with Septic Shock Phenotype,” Crit. Care Med., 2016

    2016

  48. [56]

    Machine learning models for early sepsis recognition in the neonatal intensive care unit using readily available electronic health record data.,

    A. J. Masino et al., “Machine learning models for early sepsis recognition in the neonatal intensive care unit using readily available electronic health record data.,” PLoS One, 2019. Appendix Table A1 Identifiers used in Fig.2 Identifier Distribution Volume Shape Orientation ...

    2019

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.