Blood-gas history and arterial oxygen target classification after acute brain injury: a temporal and multicentre external validation study
Journal of Intensive Care collection revision, 21 October 2026. Locations refer to the revised main manuscript (27 pages) and supplement (13 pages), with printed manuscript line numbers. Table cells are unnumbered; table locations include headings and footnotes. Figure locations identify legends; graphics are supplied separately.
The analysis protocol, code and aggregate results are provided in Additional file 2. Fitted model objects and complete reconstruction parameters are excluded from the public supplement under the source-data sharing conditions. No accessible model deposit is claimed. TRIPOD+AI items 12g and 22 are marked partial; the code and methodological description do not supply exact replay of the original fitted model.
This is a reporting map, not certification of clinical validation, data-access credentials or an institutional ethics decision. Author declarations reflect the supplied final manuscript.
| Item and topic | Status | Pages and lines | Section and evidence | Note |
|---|---|---|---|---|
| 1. Study identification | Partial | Main pp. 1–2 Main lines 1–2, 18–26 | Title; Abstract / MethodsLocation evidenceMain / Title: pp. 1; printed lines 1–2 | Main / Abstract / Methods: pp. 1–2; printed lines 18–26 | Population (acute brain injury), outcome (arterial oxygen target classification) and temporal/multicentre external validation design appear in the title; retrospective multivariable model development and evaluation are explicit in the abstract. |
| 2. Abstract checklist | Partial | Main pp. 1–2 Main lines 12–45 | Abstract / Background, Methods, Results, ConclusionsLocation evidenceMain / Abstract: pp. 1–2; printed lines 12–45 | The structured abstract contains 342 words (whitespace-delimited, excluding headings and keywords), within the Journal of Intensive Care limit of 350. See the 13-item abstract map; full retained inputs and outcome-category counts are supplied in the main text and supplement. |
| 3a. Clinical rationale | Reported | Main pp. 3–4, 21–22 Main lines 46–78, 463–484 | Background; References 1–7Location evidenceMain / Background: pp. 3–4; printed lines 46–78 | Main / References 1–7: pp. 21–22; printed lines 463–484 | The mechanical-ventilation consensus and target-range, neurocritical, ARDS ratio-monitoring and historical-estimation precedents are distinguished. |
| 3b. Population/use/users | Reported | Main pp. 3–4, 15–19 Main lines 46–78, 329–414 | Background; Discussion; ConclusionsLocation evidenceMain / Background: pp. 3–4; printed lines 46–78 | Main / Discussion: pp. 15–18; printed lines 329–405 | Main / Conclusions: pp. 18–19; printed lines 406–414 | The clinical monitoring question and research population are defined; no implemented decision tool is claimed. |
| 3c. Existing inequities | Partial | Main pp. 18 Main lines 392–405 | Discussion, limitations paragraphLocation evidenceMain / Discussion / limitations paragraph: pp. 18; printed lines 392–405 | Fairness was not assessed. Existing healthcare inequities are not developed as a separate background rationale. |
| 4. Research aims | Reported | Main pp. 4 Main lines 69–78, 80–89 | Background, final paragraph; Methods / Study design and data sourcesLocation evidenceMain / Background / final paragraph: pp. 4; printed lines 69–78 | Main / Study design and data sources: pp. 4; printed lines 80–89 | Development and locked external current-state classification objectives are stated. |
| 5a. Data provenance/representativeness | Reported | Main pp. 4, 15–18 Main lines 80–89, 329–405 | Methods / Study design and data sources; DiscussionLocation evidenceMain / Study design and data sources: pp. 4; printed lines 80–89 | Main / Discussion: pp. 15–18; printed lines 329–405 | Database versions, provenance, prior feasibility access, and selection limitations are disclosed. |
| 5b. Accrual dates | Reported | Main pp. 4, 6–8; Supplement pp. 3–4 Main lines 80–89, 129–158; Supplement lines 53–81 | Methods / Study design and data sources; Patient partitioning and model comparisons; Supplement S3Location evidenceMain / Study design and data sources: pp. 4; printed lines 80–89 | Main / Patient partitioning and model comparisons: pp. 6–8; printed lines 129–158 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Source years are distinct from conservatively inferred patient-era intervals. |
| 6a. Sites/context | Reported | Main pp. 4, 10–11 Main lines 80–89, 224–240 | Methods / Study design and data sources; Results / Cohorts and availability of blood-gas historyLocation evidenceMain / Study design and data sources: pp. 4; printed lines 80–89 | Main / Cohorts and availability of blood-gas history: pp. 10–11; printed lines 224–240 | BIDMC/Boston and United States external context are stated. Primary-pair and history-eligible hospital counts differ explicitly. |
| 6b. Eligibility rules | Reported; code review author-confirmed | Main pp. 5, 27; Supplement pp. 1 Main lines 90–110, 545–555; Supplement lines 4–22 | Methods / Participants and reference measurements; Supplement S1; Figure 1Location evidenceMain / Participants and reference measurements: pp. 5; printed lines 90–110 | Main / Figure 1. Cohort construction and evaluation partitions.: pp. 27; printed lines 545–555 | Supplement / S1. Clinical identification and analysis population: pp. 1; printed lines 4–22 | Executable operational eligibility is supplied and the clinical coauthor review of eligibility criteria, diagnostic codes and variable definitions is confirmed. This is code-definition review, not patient-level source-record adjudication. |
| 6c. Treatment handling | Partial | Main pp. 5, 15–18; Supplement pp. 1 Main lines 90–110, 329–405; Supplement lines 4–22 | Methods / Participants and reference measurements; Discussion; Supplement S1Location evidenceMain / Participants and reference measurements: pp. 5; printed lines 90–110 | Main / Discussion: pp. 15–18; printed lines 329–405 | Supplement / S1. Clinical identification and analysis population: pp. 1; printed lines 4–22 | Ventilation and treatment completeness were not required or uniformly established; treatment effects were not estimated. |
| 7. Cleaning/quality | Partial | Main pp. 5–6; Supplement pp. 2–3 Main lines 90–128; Supplement lines 23–52 | Methods / Participants and reference measurements; Measurement pairing and timestamp-faithful history; Supplement S2Location evidenceMain / Participants and reference measurements: pp. 5; printed lines 90–110 | Main / Measurement pairing and timestamp-faithful history: pp. 6; printed lines 111–128 | Supplement / S2. Specimen evidence, measurement pairing, and availability: pp. 2–3; printed lines 23–52 | Numeric, unit, duplicate and timing rules are explicit. Cleaning equivalence across demographic groups was not assessed. |
| 8a. Endpoint/timing | Reported | Main pp. 5; Supplement pp. 2–3, 8 Main lines 90–110; Supplement lines 23–52, 159–164 | Methods / Participants and reference measurements; Supplement S2; Table S2Location evidenceMain / Participants and reference measurements: pp. 5; printed lines 90–110 | Supplement / S2. Specimen evidence, measurement pairing, and availability: pp. 2–3; printed lines 23–52 | Supplement / Table S2. Measurement-level oxygen-reference distributions: pp. 8; printed lines 159–164 | Observed PaO2 categories, boundaries and differing specimen evidence are explicit. |
| 8b. Endpoint assessors | Not applicable: numeric recorded assay | Main pp. 5; Supplement pp. 2–3 Main lines 90–110; Supplement lines 23–52 | Methods / Participants and reference measurements; Supplement S2Location evidenceMain / Participants and reference measurements: pp. 5; printed lines 90–110 | Supplement / S2. Specimen evidence, measurement pairing, and availability: pp. 2–3; printed lines 23–52 | No subjective outcome assessor or independent clinical specimen-adjudication panel was used. |
| 8c. Endpoint blinding | Partial | Main pp. 4; Supplement pp. 2–3 Main lines 80–89; Supplement lines 23–52 | Methods / Study design and data sources; Supplement S2Location evidenceMain / Study design and data sources: pp. 4; printed lines 80–89 | Supplement / S2. Specimen evidence, measurement pairing, and availability: pp. 2–3; printed lines 23–52 | Labels are automated numeric categories. Prior data exposure is disclosed; a blinded human outcome assessment is not claimed. |
| 9a. Candidate selection | Reported | Main pp. 6–8; Supplement pp. 3–4 Main lines 129–158; Supplement lines 53–81 | Methods / Patient partitioning and model comparisons; Supplement S3Location evidenceMain / Patient partitioning and model comparisons: pp. 6–8; printed lines 129–158 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Small prespecified input sets; no subtype or contemporaneous blood-gas predictor selection. |
| 9b. Input definitions/timing/blinding | Reported | Main pp. 6–8, 10; Supplement pp. 2–4, 6–7 Main lines 111–158, 204–220; Supplement lines 23–81, 114–151 | Methods / Measurement pairing and timestamp-faithful history; Patient partitioning and model comparisons; Supplement S2–S3; Methods / Post hoc respiratory-support extension; Supplement S6Location evidenceMain / Measurement pairing and timestamp-faithful history: pp. 6; printed lines 111–128 | Main / Patient partitioning and model comparisons: pp. 6–8; printed lines 129–158 | Supplement / S2. Specimen evidence, measurement pairing, and availability: pp. 2–3; printed lines 23–52 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Main / Post hoc respiratory-support extension: pp. 10; printed lines 204–220 | Supplement / S6. Post hoc respiratory-support extension: pp. 6–7; printed lines 114–151 | Input definitions, transformations, timing and prior-result availability are specified. |
| 9c. Input assessors | Not applicable: structured extraction | Main pp. 6; Supplement pp. 2–3 Main lines 111–128; Supplement lines 23–52 | Methods / Measurement pairing and timestamp-faithful history; Supplement S2Location evidenceMain / Measurement pairing and timestamp-faithful history: pp. 6; printed lines 111–128 | Supplement / S2. Specimen evidence, measurement pairing, and availability: pp. 2–3; printed lines 23–52 | No subjective predictor-adjudication panel was used. |
| 10. Sample-size rationale | Reported | Main pp. 8–9 Main lines 176–203 | Methods / Statistical analysis and prespecified supplementary evaluationsLocation evidenceMain / Statistical analysis and prespecified supplementary evaluations: pp. 8–9; printed lines 176–203 | All eligible observations under frozen rules were used; absence of a formal sample-size calculation is explicit. |
| 11. Missingness handling | Reported | Main pp. 6–9, 24; Supplement pp. 2–4, 7–8, 10–11 Main lines 129–158, 176–203, 512–516; Supplement lines 23–81, 153–164, 187–192 | Methods / Patient partitioning and model comparisons; Statistical analysis and prespecified supplementary evaluations; Supplement S2–S3; Tables 1/S1/S2/S7Location evidenceMain / Patient partitioning and model comparisons: pp. 6–8; printed lines 129–158 | Main / Statistical analysis and prespecified supplementary evaluations: pp. 8–9; printed lines 176–203 | Main / Table 1. Characteristics of the primary paired cohorts: pp. 24; printed lines 512–516 | Supplement / S2. Specimen evidence, measurement pairing, and availability: pp. 2–3; printed lines 23–52 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Supplement / Table S1. Patient-level history availability and boundary-era cohort: pp. 7–8; printed lines 153–158 | Supplement / Table S2. Measurement-level oxygen-reference distributions: pp. 8; printed lines 159–164 | Supplement / Table S7. Fixed-range description and availability bookkeeping: pp. 10–11; printed lines 187–192 | Age and included oxygen values have no missing values; age-imputation rule remains documented. Prior absence and unknown/other sex are disclosed separately. |
| 12a. Data partitioning | Reported | Main pp. 6–8, 27; Supplement pp. 3–4 Main lines 129–158, 545–555; Supplement lines 53–81 | Methods / Patient partitioning and model comparisons; Supplement S3; Figure 1Location evidenceMain / Patient partitioning and model comparisons: pp. 6–8; printed lines 129–158 | Main / Figure 1. Cohort construction and evaluation partitions.: pp. 27; printed lines 545–555 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Patient-disjoint interval-based split, calibration allocation and grouped cross-validation are specified. |
| 12b. Input transformations | Reported | Main pp. 6–8, 10; Supplement pp. 3–4, 6–7 Main lines 129–158, 204–220; Supplement lines 53–81, 114–151 | Methods / Patient partitioning and model comparisons; Supplement S3; Methods / Post hoc respiratory-support extension; Supplement S6Location evidenceMain / Patient partitioning and model comparisons: pp. 6–8; printed lines 129–158 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Main / Post hoc respiratory-support extension: pp. 10; printed lines 204–220 | Supplement / S6. Post hoc respiratory-support extension: pp. 6–7; printed lines 114–151 | Hinges, logarithms, polynomial/interaction terms, age cap and standardisation are specified. |
| 12c. Fitting/tuning/validation | Reported | Main pp. 6–8, 10; Supplement pp. 3–4, 6–7 Main lines 129–158, 204–220; Supplement lines 53–81, 114–151 | Methods / Patient partitioning and model comparisons; Supplement S3; Methods / Post hoc respiratory-support extension; Supplement S6Location evidenceMain / Patient partitioning and model comparisons: pp. 6–8; printed lines 129–158 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Main / Post hoc respiratory-support extension: pp. 10; printed lines 204–220 | Supplement / S6. Post hoc respiratory-support extension: pp. 6–7; printed lines 114–151 | Fitting, regularisation grid, patient-grouped tuning, and final training are specified. |
| 12d. Cluster heterogeneity | Reported | Main pp. 8–9; Supplement pp. 5, 10 Main lines 176–203; Supplement lines 95–113, 180–186 | Methods / Statistical analysis and prespecified supplementary evaluations; Supplement S5; Table S6Location evidenceMain / Statistical analysis and prespecified supplementary evaluations: pp. 8–9; printed lines 176–203 | Supplement / S5. Uncertainty, heterogeneity, and sensitivity analyses: pp. 5; printed lines 95–113 | Supplement / Table S6. External weighting and hospital context: pp. 10; printed lines 180–186 | Patient clustering, observed-hospital scope, sparse-centre pooling and hospital weighting are addressed. |
| 12e. Evaluation measures/plots | Reported | Main pp. 8–9; Supplement pp. 4–5 Main lines 159–203; Supplement lines 82–113 | Methods / Performance measures and the in-target error gate; Statistical analysis and prespecified supplementary evaluations; Supplement S4–S5Location evidenceMain / Performance measures and the in-target error gate: pp. 8; printed lines 159–175 | Main / Statistical analysis and prespecified supplementary evaluations: pp. 8–9; printed lines 176–203 | Supplement / S4. In-target decisions and threshold selection: pp. 4–5; printed lines 82–94 | Supplement / S5. Uncertainty, heterogeneity, and sensitivity analyses: pp. 5; printed lines 95–113 | Brier contrasts, calibration, class metrics, error and both coverage denominators are defined. |
| 12f. Recalibration methods | Not performed, as prespecified | Main pp. 6–8; Supplement pp. 3–5 Main lines 129–158; Supplement lines 53–81, 95–113 | Methods / Patient partitioning and model comparisons; Supplement S3/S5Location evidenceMain / Patient partitioning and model comparisons: pp. 6–8; printed lines 129–158 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Supplement / S5. Uncertainty, heterogeneity, and sensitivity analyses: pp. 5; printed lines 95–113 | No held-out or external recalibration/model updating was performed. |
| 12g. Prediction calculation | Partial | Main pp. 20; Supplement pp. 3–4 Main lines 437–445; Supplement lines 53–81 | Supplement S3; Declarations / Availability of data and materials; Additional file 2 / reproducibility / README_reproduction.txtLocation evidenceMain / Availability of data and materials: pp. 20; printed lines 437–445 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Feature transformations, preprocessing, class order and the softmax calculation are described, and original training/evaluation code is supplied. Fitted model objects and complete reconstruction parameters are excluded from the public supplement under the stated source-data sharing conditions; no accessible model deposit is claimed. Exact replay of the authors' fitted model is therefore not provided here. |
| 13. Imbalance handling | Reported | Main pp. 6–8; Supplement pp. 3–4 Main lines 129–158; Supplement lines 53–81 | Methods / Patient partitioning and model comparisons; Supplement S3Location evidenceMain / Patient partitioning and model comparisons: pp. 6–8; printed lines 129–158 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | No class weighting or balancing was used. |
| 14. Fairness methods | Not performed; limitation reported | Main pp. 18; Supplement pp. 5 Main lines 392–405; Supplement lines 95–113 | Discussion, limitations paragraph; Supplement S5Location evidenceMain / Discussion / limitations paragraph: pp. 18; printed lines 392–405 | Supplement / S5. Uncertainty, heterogeneity, and sensitivity analyses: pp. 5; printed lines 95–113 | No fairness intervention or demographic/skin-pigmentation fairness analysis was performed. |
| 15. Outputs/thresholds | Reported | Main pp. 8, 26; Supplement pp. 4–5 Main lines 159–175, 522–528; Supplement lines 82–94 | Methods / Performance measures and the in-target error gate; Supplement S4; Table 3Location evidenceMain / Performance measures and the in-target error gate: pp. 8; printed lines 159–175 | Main / Table 3. Ordinary in-target calls and their false in-target fraction: pp. 26; printed lines 522–528 | Supplement / S4. In-target decisions and threshold selection: pp. 4–5; printed lines 82–94 | Probabilities, ordinary calls and selective calls are distinct. Null thresholds mean zero calls and undefined error. |
| 16. Dataset differences | Reported | Main pp. 4–6, 15–18, 24; Supplement pp. 7–8 Main lines 80–128, 329–405, 512–516; Supplement lines 153–164 | Methods / Study design and data sources; Participants and reference measurements; Measurement pairing and timestamp-faithful history; Tables 1/S1/S2; DiscussionLocation evidenceMain / Study design and data sources: pp. 4; printed lines 80–89 | Main / Participants and reference measurements: pp. 5; printed lines 90–110 | Main / Measurement pairing and timestamp-faithful history: pp. 6; printed lines 111–128 | Main / Table 1. Characteristics of the primary paired cohorts: pp. 24; printed lines 512–516 | Main / Discussion: pp. 15–18; printed lines 329–405 | Supplement / Table S1. Patient-level history availability and boundary-era cohort: pp. 7–8; printed lines 153–158 | Supplement / Table S2. Measurement-level oxygen-reference distributions: pp. 8; printed lines 159–164 | Source context, specimen documentation, timing, case mix and outcome distributions are compared with limitations. |
| 17. Ethics/consent | Reported; local requirement author-confirmed | Main pp. 19–20 Main lines 425–434 | Declarations / Ethics approval and consent to participateLocation evidenceMain / Ethics approval and consent to participate: pp. 19–20; printed lines 425–434 | The declaration distinguishes original database governance from the present deidentified secondary analysis. Original MIMIC review/consent-waiver and eICU deidentification are source-supported. No additional institutional review or individual consent is reported as required, based on author confirmation; no new local committee decision or approval number is claimed. |
| 18a. Financial support | Reported from author confirmation | Main pp. 20 Main lines 448–450 | Declarations / FundingLocation evidenceMain / Funding: pp. 20; printed lines 448–450 | No specific funding is reported by the author. |
| 18b. Competing interests | Reported from author confirmation | Main pp. 20 Main lines 446–447 | Declarations / Competing interestsLocation evidenceMain / Competing interests: pp. 20; printed lines 446–447 | No competing interests are reported by the author. |
| 18c. Protocol access | Reported | Main pp. 4; Supplement pp. 12–13 Main lines 80–89; Supplement lines 220–231 | Methods / Study design and data sources; Supplement / Contents of Additional file 2; Additional file 2 / reproducibilityLocation evidenceMain / Study design and data sources: pp. 4; printed lines 80–89 | Supplement / Contents of Additional file 2: pp. 12–13; printed lines 220–231 | The accompanying archive contains the internally frozen protocol and lock records. This is not public preregistration. |
| 18d. Registration status | Reported as internal nonregistered freeze | Main pp. 4 Main lines 80–89 | Methods / Study design and data sourcesLocation evidenceMain / Study design and data sources: pp. 4; printed lines 80–89 | The manuscript explicitly distinguishes computational freezing from public preregistration. |
| 18e. Data access | Reported; access author-confirmed | Main pp. 20, 22 Main lines 437–445, 485–494 | Declarations / Availability of data and materials; References 8–11Location evidenceMain / Availability of data and materials: pp. 20; printed lines 437–445 | Main / References 8–11: pp. 22; printed lines 485–494 | Source versions and PhysioNet credentialing, training and data-use conditions are stated. Authorised access for this study is author-confirmed; no credentials, certificate identifiers or patient-level records are redistributed. |
| 18f. Analysis-code access | Reported | Main pp. 20; Supplement pp. 12–13 Main lines 437–445; Supplement lines 220–231 | Declarations / Availability of data and materials; Supplement / Contents of Additional file 2; Additional file 2 / reproducibilityLocation evidenceMain / Availability of data and materials: pp. 20; printed lines 437–445 | Supplement / Contents of Additional file 2: pp. 12–13; printed lines 220–231 | Executable code is supplied in the accompanying reproducibility archive. A public repository is not claimed. |
| 19. Public involvement | Reported | Main pp. 10 Main lines 221–222 | Methods / Statistical analysis and prespecified supplementary evaluations, Patient and public involvement paragraphLocation evidenceMain / Patient and public involvement statement: pp. 10; printed lines 221–222 | The authors confirm that patients and the public were not involved in study design, conduct, reporting or dissemination. |
| 20a. Cohort flow | Reported | Main pp. 10–11, 27; Supplement pp. 7–8, 10–11 Main lines 224–240, 545–555; Supplement lines 153–164, 187–192 | Results / Cohorts and availability of blood-gas history; Figure 1; Tables S1/S2/S7; descriptives/cohort_flow_exact.csvLocation evidenceMain / Cohorts and availability of blood-gas history: pp. 10–11; printed lines 224–240 | Main / Figure 1. Cohort construction and evaluation partitions.: pp. 27; printed lines 545–555 | Supplement / Table S1. Patient-level history availability and boundary-era cohort: pp. 7–8; printed lines 153–158 | Supplement / Table S2. Measurement-level oxygen-reference distributions: pp. 8; printed lines 159–164 | Supplement / Table S7. Fixed-range description and availability bookkeeping: pp. 10–11; printed lines 187–192 | Screening/pairing/partition counts and history-availability reasons are supplied. Nonsequential patient exclusion counts may overlap. |
| 20b. Cohort characteristics/missingness | Reported with stated limits | Main pp. 24; Supplement pp. 7–8 Main lines 512–516; Supplement lines 153–164 | Table 1; Tables S1/S2; descriptives/patient_baseline_full.csv; descriptives/measurement_distributions.csvLocation evidenceMain / Table 1. Characteristics of the primary paired cohorts: pp. 24; printed lines 512–516 | Supplement / Table S1. Patient-level history availability and boundary-era cohort: pp. 7–8; printed lines 153–158 | Supplement / Table S2. Measurement-level oxygen-reference distributions: pp. 8; printed lines 159–164 | Patient and measurement summaries are distinct. Sex unknown/other count, complete selected age/oxygen inputs and omitted treatment context are disclosed. |
| 20c. Development/evaluation distributions | Reported with stated limits | Main pp. 24; Supplement pp. 7–8 Main lines 512–516; Supplement lines 153–164 | Tables 1/S1/S2; descriptives/patient_baseline_full.csv; descriptives/measurement_distributions.csvLocation evidenceMain / Table 1. Characteristics of the primary paired cohorts: pp. 24; printed lines 512–516 | Supplement / Table S1. Patient-level history availability and boundary-era cohort: pp. 7–8; printed lines 153–158 | Supplement / Table S2. Measurement-level oxygen-reference distributions: pp. 8; printed lines 159–164 | Training, calibration, later-era and external distributions are supplied; cohort summaries do not imply measured treatment equivalence. |
| 21. Analysis-specific denominators | Reported | Main pp. 10–11, 25–26; Supplement pp. 8–10 Main lines 224–240, 517–528; Supplement lines 159–164, 171–186 | Results / Cohorts and availability of blood-gas history; Tables 2/3/S2/S4/S5/S6Location evidenceMain / Cohorts and availability of blood-gas history: pp. 10–11; printed lines 224–240 | Main / Table 2. Matched probability scores and paired model contrasts: pp. 25; printed lines 517–521 | Main / Table 3. Ordinary in-target calls and their false in-target fraction: pp. 26; printed lines 522–528 | Supplement / Table S2. Measurement-level oxygen-reference distributions: pp. 8; printed lines 159–164 | Supplement / Table S4. Increment by prespecified historical blood-gas interval: pp. 9; printed lines 171–174 | Supplement / Table S5. Pairing and record-availability sensitivity analyses: pp. 10; printed lines 175–179 | Supplement / Table S6. External weighting and hospital context: pp. 10; printed lines 180–186 | Analysis-specific patient/measurement counts and three-class counts are supplied, with overlapping patient denominators explained. |
| 22. Executable model/access | Partial | Main pp. 20; Supplement pp. 3–4 Main lines 437–445; Supplement lines 53–81 | Supplement S3; Declarations / Availability of data and materials; Additional file 2 / reproducibility / README_reproduction.txtLocation evidenceMain / Availability of data and materials: pp. 20; printed lines 437–445 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Feature transformations, preprocessing, class order and the softmax calculation are described, and original training/evaluation code is supplied. Fitted model objects and complete reconstruction parameters are excluded from the public supplement under the stated source-data sharing conditions; no accessible model deposit is claimed. Exact replay of the authors' fitted model is therefore not provided here. |
| 23a. Performance/intervals/subgroups | Partial | Main pp. 11–15, 25–27; Supplement pp. 5, 8–9, 11–12 Main lines 241–279, 306–328, 517–528, 556–574; Supplement lines 95–113, 165–170, 193–202 | Results / Blood-gas history improves probability classification; In-target calls remain unreliable; Tables 2/3/S3; Figures 2/3; Supplement S5; Results / Adjustment for recorded respiratory support; Table S8Location evidenceMain / Blood-gas history improves probability classification: pp. 11–12; printed lines 241–257 | Main / In-target calls remain unreliable: pp. 12–13; printed lines 258–279 | Main / Table 2. Matched probability scores and paired model contrasts: pp. 25; printed lines 517–521 | Main / Table 3. Ordinary in-target calls and their false in-target fraction: pp. 26; printed lines 522–528 | Main / Figure 2. Probability-classification increment and the error–coverage relationship.: pp. 27; printed lines 556–567 | Main / Figure 3. External calibration of the three oxygen-reference categories.: pp. 27; printed lines 568–574 | Supplement / Table S3. External class-specific score, discrimination and calibration: pp. 8–9; printed lines 165–170 | Supplement / S5. Uncertainty, heterogeneity, and sensitivity analyses: pp. 5; printed lines 95–113 | Main / Adjustment for recorded respiratory support: pp. 14–15; printed lines 306–328 | Supplement / Table S8. Post hoc respiratory-support extension: matched scores and paired contrasts: pp. 11–12; printed lines 193–202 | Primary contrasts and key fractions have cluster intervals. AUROC/calibration coefficients remain descriptive point estimates; fairness performance was not assessed. |
| 23b. Between-cluster results | Reported | Main pp. 13; Supplement pp. 10, 12 Main lines 280–292; Supplement lines 180–186, 212–219 | Results / Durability across information age and hospitals; Table S6; Figure S2Location evidenceMain / Durability across information age and hospitals: pp. 13; printed lines 280–292 | Supplement / Table S6. External weighting and hospital context: pp. 10; printed lines 180–186 | Supplement / Figure S2. Hospital-level variation in the historical-information increment.: pp. 12; printed lines 212–219 | All centres retained; adequately sized centres displayed individually and smaller centres pooled. |
| 24. Updating results | Not performed, as prespecified | Main pp. 6–8, 15–18; Supplement pp. 3–4 Main lines 129–158, 329–405; Supplement lines 53–81 | Methods / Patient partitioning and model comparisons; Supplement S3; DiscussionLocation evidenceMain / Patient partitioning and model comparisons: pp. 6–8; printed lines 129–158 | Main / Discussion: pp. 15–18; printed lines 329–405 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | No external model updating or recalibration results are claimed. |
| 25. Contextual interpretation | Reported | Main pp. 15–19 Main lines 329–414 | Discussion; ConclusionsLocation evidenceMain / Discussion: pp. 15–18; printed lines 329–405 | Main / Conclusions: pp. 18–19; printed lines 406–414 | The comparison uses a limited three-input current-information model; a post hoc extension locked before fitting adds recorded FiO2, PEEP and ventilation status, with ventilation status unknown for 71.8% of external measurements. Average increment, external miscalibration, the failed low-error criterion, prior studies and unassessed fairness remain distinct. |
| 26. Bias/uncertainty/generalizability | Reported | Main pp. 4, 18 Main lines 80–89, 392–405 | Discussion, limitations paragraph; Methods / Study design and data sourcesLocation evidenceMain / Discussion / limitations paragraph: pp. 18; printed lines 392–405 | Main / Study design and data sources: pp. 4; printed lines 80–89 | Sampling, coding, specimen evidence, timing, small subsets and conditional uncertainty are acknowledged in the limitations paragraph; prior feasibility inspection of both databases is disclosed in Methods. |
| 27a. Unavailable/poor inputs | Reported for research inputs | Main pp. 15–19; Supplement pp. 2–4 Main lines 329–414; Supplement lines 23–81 | Supplement S2–S3; Discussion; ConclusionsLocation evidenceMain / Discussion: pp. 15–18; printed lines 329–405 | Main / Conclusions: pp. 18–19; printed lines 406–414 | Supplement / S2. Specimen evidence, measurement pairing, and availability: pp. 2–3; printed lines 23–52 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Unavailable oxygen/history inputs prevent the corresponding research classification. Missing age rule is specified; deployment behaviour was not tested. |
| 27b. User interaction/expertise | Not assessed: no deployment | Main pp. 15–19 Main lines 329–414 | Discussion; ConclusionsLocation evidenceMain / Discussion: pp. 15–18; printed lines 329–405 | Main / Conclusions: pp. 18–19; printed lines 406–414 | No clinical interface, user study or deployed workflow exists for these research models. |
| 27c. Future applicability | Reported | Main pp. 15–19 Main lines 329–414 | Discussion; ConclusionsLocation evidenceMain / Discussion: pp. 15–18; printed lines 329–405 | Main / Conclusions: pp. 18–19; printed lines 406–414 | The manuscript does not support replacing current blood gases, changing treatment targets or setting a sampling interval. |
| Item and topic | Status | Pages and lines | Section and evidence | Note |
|---|---|---|---|---|
| 1. Model/population/outcome title | Partial | Main pp. 1 Main lines 1–2 | TitleLocation evidenceMain / Title: pp. 1; printed lines 1–2 | Population, oxygen-target outcome and validation design are explicit; the title does not explicitly state multivariable model development. |
| 2. Clinical rationale | Reported | Main pp. 1 Main lines 13–17 | Abstract / BackgroundLocation evidenceMain / Abstract / Background: pp. 1; printed lines 13–17 | Clinical monitoring rationale is clear. |
| 3. Development/evaluation aims | Reported | Main pp. 1–2 Main lines 12–45 | Abstract / Background and MethodsLocation evidenceMain / Abstract: pp. 1–2; printed lines 12–45 | Incremental classification question and model development/evaluation are stated. |
| 4. Data sources | Reported | Main pp. 1–2 Main lines 18–26 | Abstract / MethodsLocation evidenceMain / Abstract / Methods: pp. 1–2; printed lines 18–26 | MIMIC-IV 3.1 and eICU 2.0 identified. |
| 5. Eligibility/setting | Partial | Main pp. 1–2 Main lines 12–45 | Abstract / Background and MethodsLocation evidenceMain / Abstract: pp. 1–2; printed lines 12–45 | Adults with ABI, the retrospective design and the first 72 ICU hours are stated. Detailed specimen/pairing eligibility and geography are in main Methods. |
| 6. Endpoint/horizon | Reported | Main pp. 1–2 Main lines 18–26 | Abstract / MethodsLocation evidenceMain / Abstract / Methods: pp. 1–2; printed lines 18–26 | The three PaO2 categories (<80, 80–120, >120 mmHg) and the consensus origin of the middle range are stated. This is current-state classification rather than a future prognostic horizon. |
| 7. Model building/validation | Partial | Main pp. 1–2 Main lines 18–26 | Abstract / MethodsLocation evidenceMain / Abstract / Methods: pp. 1–2; printed lines 18–26 | Penalised multinomial models and patient-disjoint later-era/external evaluation stated; grouped tuning details remain in main Methods/S3. |
| 8. Performance measures | Partial | Main pp. 1–2 Main lines 12–45 | Abstract / Methods and ResultsLocation evidenceMain / Abstract: pp. 1–2; printed lines 12–45 | The primary Brier contrast with its 95% CI, the false in-target error and the prespecified 5% error gate are included; the multiclass definition, patient-cluster bootstrap, coverage and calibration measures are reported in main Methods and Results. |
| 9. Patients/outcomes | Partial | Main pp. 2 Main lines 27–37 | Abstract / ResultsLocation evidenceMain / Abstract / Results: pp. 2; printed lines 27–37 | Patient and measurement counts are provided; three outcome-category counts are in Table S2, not the abstract. |
| 10. Final inputs | Partial | Main pp. 1–2 Main lines 18–26 | Abstract / MethodsLocation evidenceMain / Abstract / Methods: pp. 1–2; printed lines 18–26 | B inputs are explicit: current SpO2, age and elapsed ICU time. The prior-only and current-plus-prior input groups are summarised; prior paired SpO2 and full transformations are detailed in main Methods and Supplement S3. |
| 11. Performance/intervals | Reported for primary contrasts | Main pp. 2 Main lines 27–37 | Abstract / ResultsLocation evidenceMain / Abstract / Results: pp. 2; printed lines 27–37 | Brier point estimates and primary/secondary contrast confidence intervals are reported. Classification fractions are point estimates in the abstract. |
| 12. Interpretation | Reported | Main pp. 2 Main lines 38–45 | Abstract / ConclusionsLocation evidenceMain / Abstract / Conclusions: pp. 2; printed lines 38–45 | The increment, the failure of in-target calls and the unmet 5% error gate are explicit; external miscalibration is reported in main Results. Recommendations concern information display, arterial confirmation and validation standards; no bedside readiness claim is made. |
| 13. Registry/repository | Not in abstract; no public registration | Main pp. 4 Main lines 80–89 | Main Methods / Study design and data sourcesLocation evidenceMain / Study design and data sources: pp. 4; printed lines 80–89 | The main text explicitly reports no public preregistration. The internally frozen protocol is supplied in Additional file 2; no registry identifier is claimed. |
| Item and topic | Status | Pages and lines | Section and evidence | Note |
|---|---|---|---|---|
| 1. Design and summary | Reported | Main pp. 1–2 Main lines 1–2, 12–45 | Title; AbstractLocation evidenceMain / Title: pp. 1; printed lines 1–2 | Main / Abstract: pp. 1–2; printed lines 12–45 | Retrospective multicentre design and balanced findings are stated. |
| 2. Scientific context | Reported | Main pp. 3–4, 21–22 Main lines 46–78, 463–484 | Background; References 1–7Location evidenceMain / Background: pp. 3–4; printed lines 46–78 | Main / References 1–7: pp. 21–22; printed lines 463–484 | Clinical reference-target context and the closest range-classification, neurocritical, ARDS ratio-monitoring and historical-estimation studies are compared. |
| 3. Objectives | Reported | Main pp. 4 Main lines 69–78 | Background, final paragraphLocation evidenceMain / Background / final paragraph: pp. 4; printed lines 69–78 | Incremental classification and external evaluation objectives stated. |
| 4. Design structure | Reported | Main pp. 4 Main lines 80–89 | Methods / Study design and data sourcesLocation evidenceMain / Study design and data sources: pp. 4; printed lines 80–89 | Retrospective current-state task and computational locks specified. |
| 5. Place and dates | Reported | Main pp. 4; Supplement pp. 3–4 Main lines 80–89; Supplement lines 53–81 | Methods / Study design and data sources; Supplement S3Location evidenceMain / Study design and data sources: pp. 4; printed lines 80–89 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Database geography, years and inferred era split distinguished. |
| 6. Participant selection | Reported; code review author-confirmed | Main pp. 5; Supplement pp. 1–3 Main lines 90–110; Supplement lines 4–52 | Methods / Participants and reference measurements; Supplement S1/S2Location evidenceMain / Participants and reference measurements: pp. 5; printed lines 90–110 | Supplement / S1. Clinical identification and analysis population: pp. 1; printed lines 4–22 | Supplement / S2. Specimen evidence, measurement pairing, and availability: pp. 2–3; printed lines 23–52 | Operational coding and pairing eligibility are supplied; a clinical coauthor reviewed these definitions. No independent patient-level clinical adjudication is claimed. This is not an exposure-matched causal cohort. |
| 7. Variable definitions | Reported | Main pp. 5–8; Supplement pp. 2–4 Main lines 90–110, 129–158; Supplement lines 23–81 | Methods / Participants and reference measurements; Patient partitioning and model comparisons; Supplement S2/S3Location evidenceMain / Participants and reference measurements: pp. 5; printed lines 90–110 | Main / Patient partitioning and model comparisons: pp. 6–8; printed lines 129–158 | Supplement / S2. Specimen evidence, measurement pairing, and availability: pp. 2–3; printed lines 23–52 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | PaO2 targets and predictor transforms defined; no causal confounder set claimed. |
| 8. Measurements | Reported with limits | Main pp. 6; Supplement pp. 2–3 Main lines 111–128; Supplement lines 23–52 | Methods / Measurement pairing and timestamp-faithful history; Supplement S2Location evidenceMain / Measurement pairing and timestamp-faithful history: pp. 6; printed lines 111–128 | Supplement / S2. Specimen evidence, measurement pairing, and availability: pp. 2–3; printed lines 23–52 | Source comparability and arterial-documentation asymmetry disclosed. |
| 9. Bias handling | Reported | Main pp. 6, 15–18; Supplement pp. 2–5 Main lines 111–128, 329–405; Supplement lines 23–81, 95–113 | Methods / Measurement pairing and timestamp-faithful history; Supplement S2/S3/S5; DiscussionLocation evidenceMain / Measurement pairing and timestamp-faithful history: pp. 6; printed lines 111–128 | Main / Discussion: pp. 15–18; printed lines 329–405 | Supplement / S2. Specimen evidence, measurement pairing, and availability: pp. 2–3; printed lines 23–52 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Supplement / S5. Uncertainty, heterogeneity, and sensitivity analyses: pp. 5; printed lines 95–113 | Leakage precautions, fixed analyses and major residual biases described. |
| 10. Study-size basis | Reported | Main pp. 8–9 Main lines 176–203 | Methods / Statistical analysis and prespecified supplementary evaluationsLocation evidenceMain / Statistical analysis and prespecified supplementary evaluations: pp. 8–9; printed lines 176–203 | All eligible bounded observations used; no formal sample-size calculation. |
| 11. Numeric variables | Reported | Main pp. 6–8; Supplement pp. 3–5 Main lines 129–158; Supplement lines 53–81, 95–113 | Methods / Patient partitioning and model comparisons; Supplement S3/S5Location evidenceMain / Patient partitioning and model comparisons: pp. 6–8; printed lines 129–158 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Supplement / S5. Uncertainty, heterogeneity, and sensitivity analyses: pp. 5; printed lines 95–113 | Transformations, thresholds and historical-time strata given. |
| 12. Analytical procedures | Reported for task | Main pp. 8–10; Supplement pp. 3–7 Main lines 176–220; Supplement lines 53–151 | Methods / Statistical analysis and prespecified supplementary evaluations; Supplement S3–S5; Methods / Post hoc respiratory-support extension; Supplement S6Location evidenceMain / Statistical analysis and prespecified supplementary evaluations: pp. 8–9; printed lines 176–203 | Supplement / S3. Partitioning, model specification, and reproducibility: pp. 3–4; printed lines 53–81 | Supplement / S4. In-target decisions and threshold selection: pp. 4–5; printed lines 82–94 | Supplement / S5. Uncertainty, heterogeneity, and sensitivity analyses: pp. 5; printed lines 95–113 | Main / Post hoc respiratory-support extension: pp. 10; printed lines 204–220 | Supplement / S6. Post hoc respiratory-support extension: pp. 6–7; printed lines 114–151 | Patient grouping, missing inputs, sensitivity analyses and descriptive subtypes addressed; loss to future follow-up is not this endpoint. |
| 13. Flow and exclusions | Reported | Main pp. 10–11, 27; Supplement pp. 8, 10–11 Main lines 224–240, 545–555; Supplement lines 159–164, 187–192 | Figure 1; Results / Cohorts and availability of blood-gas history; Tables S2/S7; descriptives/cohort_flow_exact.csvLocation evidenceMain / Figure 1. Cohort construction and evaluation partitions.: pp. 27; printed lines 545–555 | Main / Cohorts and availability of blood-gas history: pp. 10–11; printed lines 224–240 | Supplement / Table S2. Measurement-level oxygen-reference distributions: pp. 8; printed lines 159–164 | Supplement / Table S7. Fixed-range description and availability bookkeeping: pp. 10–11; printed lines 187–192 | Counts and reasons supplied; overlapping patient branches are not summed. |
| 14. Baseline and missing data | Reported with limits | Main pp. 24; Supplement pp. 7–8 Main lines 512–516; Supplement lines 153–164 | Table 1; Tables S1/S2; descriptives/patient_baseline_full.csv; descriptives/measurement_distributions.csvLocation evidenceMain / Table 1. Characteristics of the primary paired cohorts: pp. 24; printed lines 512–516 | Supplement / Table S1. Patient-level history availability and boundary-era cohort: pp. 7–8; printed lines 153–158 | Supplement / Table S2. Measurement-level oxygen-reference distributions: pp. 8; printed lines 159–164 | Included demographics and missingness disclosed; treatment/neurological severity unconfirmed. No whole-stay exposure follow-up estimate. |
| 15. Observed endpoints | Reported | Supplement pp. 8 Supplement lines 159–164 | Table S2Location evidenceSupplement / Table S2. Measurement-level oxygen-reference distributions: pp. 8; printed lines 159–164 | Three PaO2-category counts supplied by partition and history eligibility. |
| 16. Primary estimates | Reported for task | Main pp. 11–13, 25–26 Main lines 241–279, 517–528 | Tables 2/3; Results / Blood-gas history improves probability classification; In-target calls remain unreliableLocation evidenceMain / Table 2. Matched probability scores and paired model contrasts: pp. 25; printed lines 517–521 | Main / Table 3. Ordinary in-target calls and their false in-target fraction: pp. 26; printed lines 522–528 | Main / Blood-gas history improves probability classification: pp. 11–12; printed lines 241–257 | Main / In-target calls remain unreliable: pp. 12–13; printed lines 258–279 | Paired scores, absolute error/coverage and precision given; relative-risk conversion not applicable. |
| 17. Other analyses | Reported | Main pp. 14–15; Supplement pp. 5, 8–12 Main lines 306–328; Supplement lines 95–113, 165–202, 204–219 | Tables S3–S7; Figures S1/S2; Supplement S5; Results / Adjustment for recorded respiratory support; Table S8Location evidenceSupplement / Table S3. External class-specific score, discrimination and calibration: pp. 8–9; printed lines 165–170 | Supplement / Table S4. Increment by prespecified historical blood-gas interval: pp. 9; printed lines 171–174 | Supplement / Table S5. Pairing and record-availability sensitivity analyses: pp. 10; printed lines 175–179 | Supplement / Table S6. External weighting and hospital context: pp. 10; printed lines 180–186 | Supplement / Table S7. Fixed-range description and availability bookkeeping: pp. 10–11; printed lines 187–192 | Supplement / Figure S1. Historical-information age and incremental probability-classification performance.: pp. 12; printed lines 204–211 | Supplement / Figure S2. Hospital-level variation in the historical-information increment.: pp. 12; printed lines 212–219 | Supplement / S5. Uncertainty, heterogeneity, and sensitivity analyses: pp. 5; printed lines 95–113 | Main / Adjustment for recorded respiratory support: pp. 14–15; printed lines 306–328 | Supplement / Table S8. Post hoc respiratory-support extension: matched scores and paired contrasts: pp. 11–12; printed lines 193–202 | Sensitivity, time, weighting and hospital analyses reported as descriptive. |
| 18. Main findings | Reported | Main pp. 15–16 Main lines 330–350 | Discussion, opening paragraphsLocation evidenceMain / Discussion / opening paragraphs: pp. 15–16; printed lines 330–350 | Findings tied to the prespecified incremental question. |
| 19. Study limitations | Reported | Main pp. 18 Main lines 392–405 | Discussion, limitations paragraphLocation evidenceMain / Discussion / limitations paragraph: pp. 18; printed lines 392–405 | Bias and precision limits acknowledged; residual bias magnitude is not quantifiable from these data. |
| 20. Interpretation | Reported | Main pp. 15–19 Main lines 329–414 | Discussion; ConclusionsLocation evidenceMain / Discussion: pp. 15–18; printed lines 329–405 | Main / Conclusions: pp. 18–19; printed lines 406–414 | External increment and clinical-use limitations are separated; no causal inference. |
| 21. Transportability | Reported | Main pp. 15–19 Main lines 329–414 | Discussion; ConclusionsLocation evidenceMain / Discussion: pp. 15–18; printed lines 329–405 | Main / Conclusions: pp. 18–19; printed lines 406–414 | Restricted to tested/pairable coded candidates and observed centres; no universal bedside validity. |
| 22. Financial support | Reported for this study; original-resource details by citation | Main pp. 20, 22 Main lines 448–450, 485–494 | Declarations / Funding; References 8–11Location evidenceMain / Funding: pp. 20; printed lines 448–450 | Main / References 8–11: pp. 22; printed lines 485–494 | Author reports no specific study funding; source-resource publications identify their own support. |
Page and printed-line locations were re-derived by matching every DOCX paragraph to the rendered PDF of the Journal of Intensive Care revision. These are manuscript-working locations and may differ from journal-production pagination.