Clinical Characteristics and Laboratory Biomarkers in ICU-admitted Septic Patients with and without Bacteremia

Sangwon Baek; Seung Jun Lee

arxiv: 2311.08433 · v3 · submitted 2023-11-14 · 🧬 q-bio.QM · cs.LG· stat.AP

Clinical Characteristics and Laboratory Biomarkers in ICU-admitted Septic Patients with and without Bacteremia

Sangwon Baek , Seung Jun Lee This is my paper

Pith reviewed 2026-05-24 05:17 UTC · model grok-4.3

classification 🧬 q-bio.QM cs.LGstat.AP

keywords bacteremiasepsisICUprocalcitoninlogistic regressionbiomarkersprediction modelmortality

0 comments

The pith

A seven-variable logistic model predicts bacteremia among septic ICU patients with AUC 0.907.

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

The study tests whether routine lab values can distinguish true bacteremia from other causes of sepsis in ICU patients. Procalcitonin alone separates the groups with an AUC of 0.845, yet adding bilirubin, neutrophil-lymphocyte ratio, platelet count, lactic acid, erythrocyte sedimentation rate, and Glasgow Coma Scale score raises performance to 0.907 in a multivariable logistic regression. The authors also note that confirmed bacteremia correlates with higher mortality. A reader would care because earlier identification could limit unnecessary broad antibiotics while focusing treatment on patients who need it.

Core claim

In 218 adult septic patients admitted to one ICU in 2019, 48 had culture-confirmed bacteremia. Procalcitonin and C-reactive protein each showed useful discrimination, but the multivariable logistic regression that combined procalcitonin, bilirubin, neutrophil-lymphocyte ratio, platelets, lactic acid, erythrocyte sedimentation rate, and Glasgow Coma Scale score produced an AUC of 0.907. Bacteremia was further associated with increased mortality on survival analysis.

What carries the argument

The multivariable logistic regression model fitted on the seven selected laboratory and clinical variables.

If this is right

The combined model separates bacteremic from non-bacteremic sepsis more accurately than procalcitonin alone.
Confirmed bacteremia tracks with elevated mortality in this population.
Routine clinical data already collected at admission can be reused for the prediction without new tests.
C-reactive protein alone performs less well than the full multivariable set.

Where Pith is reading between the lines

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

If the model holds on new data, it could shorten the interval between ICU arrival and targeted therapy decisions.
The same variables might be tested for predicting other culture-positive infections in critical care.
Coefficient stability would need checking across different hospital settings and sepsis definitions.

Load-bearing premise

The 48 bacteremia and 170 non-bacteremia cases drawn from a single 2019 ICU cohort are representative enough for the fitted coefficients to apply elsewhere.

What would settle it

Fitting the same seven-variable model to blood-culture results from an independent multi-center cohort of septic ICU patients and checking whether the AUC stays above 0.85.

read the original abstract

Few studies have investigated the diagnostic utilities of biomarkers for predicting bacteremia among septic patients admitted to intensive care units (ICU). Therefore, this study evaluated the prediction power of laboratory biomarkers to utilize those markers with high performance to optimize the predictive model for bacteremia. This retrospective cross-sectional study was conducted at the ICU department of Gyeongsang National University Changwon Hospital in 2019. Adult patients qualifying SEPSIS-3 (increase in sequential organ failure score greater than or equal to 2) criteria with at least two sets of blood culture were selected. Collected data was initially analyzed independently to identify the significant predictors, which was then used to build the multivariable logistic regression (MLR) model. A total of 218 patients with 48 cases of true bacteremia were analyzed in this research. Both CRP and PCT showed a substantial area under the curve (AUC) value for discriminating bacteremia among septic patients (0.757 and 0.845, respectively). To further enhance the predictive accuracy, we combined PCT, bilirubin, neutrophil lymphocyte ratio (NLR), platelets, lactic acid, erythrocyte sedimentation rate (ESR), and Glasgow Coma Scale (GCS) score to build the predictive model with an AUC of 0.907 (95% CI, 0.843 to 0.956). In addition, a high association between bacteremia and mortality rate was discovered through the survival analysis (0.004). While PCT is certainly a useful index for distinguishing patients with and without bacteremia by itself, our MLR model indicates that the accuracy of bacteremia prediction substantially improves by the combined use of PCT, bilirubin, NLR, platelets, lactic acid, ESR, and GCS score.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Standard single-center logistic regression on sepsis biomarkers with no validation, so the AUC 0.907 is likely optimistic and the work adds little beyond existing PCT studies.

read the letter

This is a 2019 single-center retrospective look at 218 septic ICU patients (48 with bacteremia) that fits a seven-variable logistic regression and reports an AUC of 0.907 for the combination of PCT, bilirubin, NLR, platelets, lactic acid, ESR, and GCS. PCT alone reached 0.845. They also ran a survival analysis showing a mortality link (p=0.004). That's the whole result set from the abstract. The modeling is conventional multivariable logistic regression after univariate screening. Nothing in the approach or the marker list is new; similar PCT-focused models for bacteremia already appear in the literature the abstract cites. The paper does report a confidence interval around the combined AUC and flags the mortality association, which is at least transparent on those points. The clear limitation is the event count. Forty-eight positives for seven predictors sits at or below the usual 10-events-per-predictor threshold, and the abstract supplies no information on variable selection, regularization, missing-data handling, or any form of internal validation. The AUC therefore reflects in-sample performance on the same data used to estimate the coefficients. Single-hospital data from one year adds the usual generalizability concern. This is the sort of incremental local-cohort analysis that might interest a critical-care group running their own biomarker audit, but it does not change how we think about bacteremia prediction or supply reproducible coefficients. The methodological gaps are too large for the modest claim. I would not send it to peer review; the combination of low novelty and unexamined modeling choices makes it a desk reject in most venues.

Referee Report

3 major / 2 minor

Summary. The manuscript reports a retrospective analysis of 218 septic ICU patients (48 with true bacteremia) from a single 2019 cohort. It claims that CRP and PCT individually discriminate bacteremia with AUCs of 0.757 and 0.845, respectively, and that a seven-predictor multivariable logistic regression (MLR) model using PCT, bilirubin, NLR, platelets, lactic acid, ESR, and GCS achieves an AUC of 0.907 (95% CI 0.843–0.956). It further reports a significant association between bacteremia and mortality via survival analysis (p=0.004).

Significance. If the AUC claim were shown to be robust under internal validation and external testing, the combined biomarker model could meaningfully improve early bacteremia prediction in sepsis and support more targeted antibiotic use. The work adds to existing literature on PCT and related markers but currently offers no evidence that the reported performance generalizes beyond the fitting cohort.

major comments (3)

[Abstract] Abstract: the AUC of 0.907 is obtained by fitting seven coefficients to the identical 218-patient dataset (48 events) with no mention of variable selection procedure, internal validation (cross-validation, bootstrap, or split-sample), or calibration; the performance metric is therefore not independent of the model-fitting process.
[Abstract] Abstract: with only 48 positive events for a 7-predictor logistic regression the analysis falls below the conventional 10-events-per-predictor guideline, creating a direct risk of optimistic bias in the reported AUC and coefficient estimates.
[Abstract] Abstract: the survival-analysis result (p=0.004) linking bacteremia to mortality supplies no information on the statistical method, follow-up time, censoring, or adjustment for confounders, rendering this secondary claim uninterpretable.

minor comments (2)

[Abstract] Abstract: the design is labeled both 'retrospective cross-sectional' and as including survival analysis; clarify how longitudinal mortality data were obtained within a cross-sectional frame.
[Abstract] Abstract: no information is given on missing-data handling or the exact criteria used to retain the seven predictors in the final MLR model.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We respond point by point to the major comments and indicate planned revisions.

read point-by-point responses

Referee: [Abstract] Abstract: the AUC of 0.907 is obtained by fitting seven coefficients to the identical 218-patient dataset (48 events) with no mention of variable selection procedure, internal validation (cross-validation, bootstrap, or split-sample), or calibration; the performance metric is therefore not independent of the model-fitting process.

Authors: We agree that the reported AUC is apparent performance without internal validation or calibration. In revision we will add bootstrap internal validation with optimism-corrected AUC and calibration plots. revision: yes
Referee: [Abstract] Abstract: with only 48 positive events for a 7-predictor logistic regression the analysis falls below the conventional 10-events-per-predictor guideline, creating a direct risk of optimistic bias in the reported AUC and coefficient estimates.

Authors: The referee correctly notes the events-per-predictor ratio of ~6.9. This is a limitation of the single-center sample. We will discuss the risk of overfitting, report sensitivity analyses with fewer predictors or penalized regression, and note this as a study limitation. revision: partial
Referee: [Abstract] Abstract: the survival-analysis result (p=0.004) linking bacteremia to mortality supplies no information on the statistical method, follow-up time, censoring, or adjustment for confounders, rendering this secondary claim uninterpretable.

Authors: We agree the abstract omits these details. The manuscript uses Kaplan-Meier with log-rank test; we will expand the abstract and methods to specify follow-up duration, censoring, and any confounder adjustment. revision: yes

Circularity Check

1 steps flagged

In-sample AUC 0.907 reported as 'predictive' performance for 7-predictor MLR fitted on the identical 218-patient cohort

specific steps

fitted input called prediction [abstract]
"we combined PCT, bilirubin, neutrophil lymphocyte ratio (NLR), platelets, lactic acid, erythrocyte sedimentation rate (ESR), and Glasgow Coma Scale (GCS) score to build the predictive model with an AUC of 0.907 (95% CI, 0.843 to 0.956)"

Coefficients for the seven predictors are estimated from the 218-patient dataset; the AUC is then computed on that same dataset. With no validation procedure stated, the quoted 'predictive' AUC reduces directly to the in-sample goodness-of-fit of the fitted model.

full rationale

The paper's central claim is that combining seven biomarkers into an MLR yields AUC 0.907 for bacteremia prediction. Data collection, predictor selection, coefficient fitting, and AUC calculation all occur on the single 2019 ICU cohort of 218 patients (48 events). No train/test split, cross-validation, bootstrap, or external validation is described, so the reported AUC is the in-sample fit statistic rather than an independent prediction. This matches the 'fitted input called prediction' pattern exactly; the performance number is statistically forced by the fitting step itself. No other circularity patterns (self-definition, self-citation chains, ansatz smuggling, etc.) are present in the available text.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the fitted logistic regression coefficients derived from the 218-patient sample and on the assumption that blood-culture results constitute an error-free gold standard for bacteremia.

free parameters (1)

logistic regression coefficients for the seven predictors
Estimated from the 218-patient dataset to produce the reported AUC of 0.907

axioms (2)

domain assumption Logistic regression assumptions (linearity of logit, no perfect multicollinearity, independent observations) hold for these clinical variables
Invoked implicitly by the choice of MLR without reported diagnostics
domain assumption Blood culture results accurately classify true bacteremia versus contamination or false negatives
Used as the binary outcome throughout the analysis

pith-pipeline@v0.9.0 · 5822 in / 1403 out tokens · 24877 ms · 2026-05-24T05:17:28.009359+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

our MLR model indicates that the accuracy of bacteremia prediction substantially improves by the combined use of PCT, bilirubin, NLR, platelets, lactic acid, ESR, and GCS score (AUC 0.907)

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.