{"id":"9991947c-3703-4af2-a2cf-6c7f9c29d6c9","arxiv_id":"2501.09640","paper_version":2,"verdict":"UNVERDICTED","confidence":"HIGH","novelty_score":1.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A review chapter explaining EHR history, ICD coding, and MIMIC-III querying, with no new scientific findings.","lead":"This preprint is a teaching chapter that reviews electronic health records, ICD disease coding, and the MIMIC-III critical care database, including example SQL queries. It contains no new research results, and its value is limited to introductory education for healthcare data science.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 3.3 misreports MIMIC-III counts: 53,423 is ICU stays, not hospital admissions, and the 'aged 60 and above' claim is unsupported; this undercuts the chapter's central schema-semantics lesson.","rationale":"The reader's weakest assumption correctly identifies the internal statistical inconsistency in Section 3.3. My review sharpens it: the stated 53,000+ figure is not merely an unexplained discrepancy but likely confuses ICU stays with hospital admissions, which is exactly the identifier-conflation error the chapter warns against. This is load-bearing because the chapter's educational value depends on accurate schema semantics and reliable reporting of MIMIC-III counts. The concern does not change the overall verdict: the work is educational review material rather than a novel scientific claim, and the Pith verdict of UNVERDICTED already reflects that. I found no separate, more fundamental flaw; the SQL walkthroughs and entity-relationship exposition are potentially useful, but the Section 3.3 error should be corrected before the chapter is used as a reliable reference.","tokens_in":15258,"tokens_out":3102,"duration_ms":31395,"concrete_test":"Run the canonical MIMIC-III v1.4 counts directly against the database: SELECT COUNT(DISTINCT SUBJECT_ID) FROM PATIENTS; SELECT COUNT(DISTINCT HADM_ID) FROM ADMISSIONS; SELECT COUNT(DISTINCT ICUSTAY_ID) FROM ICUSTAYS; and the hospital-admission count restricted to patients aged 60 or older at admission. If the results are 38,597, 49,785, 53,423, and a number below 53,000, then Section 3.3's 'over 53,000 distinct hospital admissions for patients aged 60 years and above' is refuted by the database itself. Cross-check the chapter's own Figures 9 and 10 to confirm the internal contradiction.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The chapter's central pedagogical claim is that correct MIMIC-III data extraction requires understanding the schema because different identifiers yield different counts. The most load-bearing assumption is therefore that the chapter's own reported MIMIC-III statistics are accurate. That assumption fails in Section 3.3, which states that 'the MIMIC-III database contains over 53,000 distinct hospital admissions for patients aged 60 years and above.' This is contradicted by the chapter's own Figures 9 and 10, which report 49,785 total hospital admissions and 53,423 ICU stays. The publicly documented MIMIC-III v1.4 counts are 38,597 distinct adult patients, 49,785 hospital admissions, and 53,423 ICU stays; 53,423 is the ICU-stay count, not a hospital-admission count, and the sentence adds an age restriction that is not part of that figure. Because the chapter explicitly warns that conflating identifiers such as hospital admissions and ICU stays leads to subtly different results, this error is not a cosmetic typo: it is the same class of mistake the chapter tells readers to avoid, and it undermines confidence in the factual reliability of the case studies and statistics used throughout.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This preprint is an expository chapter on electronic health records and digital twins, with a detailed tutorial on the MIMIC-III critical care database. It reviews the transition to EHRs in the UK and USA, the ICD-9/ICD-10 coding systems, MIMIC-III's relational schema, descriptive statistics queries, mortality and length-of-stay outcomes, vital sign extraction, ICD-9 code distribution, clinical concepts, and a catheterization mortality case study, before discussing the path from descriptive to prescriptive analytics and digital twins. The central pedagogical claim is that correct analysis of MIMIC-III requires understanding the schema because different identifiers (patients, hospital admissions, ICU stays) yield different counts.","tokens_in":15499,"tokens_out":6784,"duration_ms":62977,"significance":"If the factual issues are corrected, this chapter would be a useful pedagogical resource for researchers entering critical-care data analytics. Its strengths are the clear entity-relationship exposition, the emphasis on identifier semantics and the consequences of query choices, the practical SQL query examples (presented as figures), and the accessible treatment of ICD coding and clinical concepts. The chapter does not present a new method or derivation, and it does not ship code or machine-checked proofs; its value is tutorial rather than research.","major_comments":[{"comment":"The sentence 'The MIMIC-III database contains over 53,000 distinct hospital admissions for patients aged 60 years and above' is internally inconsistent with the chapter's own Figures 9 and 10, which report 49,785 hospital admissions and 53,423 ICU stays, and it conflicts with published MIMIC-III v1.4 counts. The number 53,423 is the total count of ICU stays, not hospital admissions, and the age restriction is unsupported by the cited figures. Because the chapter's central lesson is that conflating hospital admission and ICU stay identifiers produces different results, this error reproduces the very mistake the text warns against and should be corrected (e.g., to 'the database contains 49,785 hospital admissions and 53,423 ICU stays; the later case studies often restrict to patients aged 60 and above').","section":"3.3"},{"comment":"The ICU mortality paragraph states that the query 'focuses on patients older than 60 years' but then reports 'total ICU mortality is about 8.5%, corresponding to approximately 4,565 patients.' Since 8.5% of 53,423 is about 4,541, the reported mortality count matches the all-adult ICU population rather than a >60-year-old subset. The authors should either provide the cohort size for the age-restricted query or clarify that the 8.5% figure is for all adults; as written, the age filter and the reported numbers are inconsistent and undermine the schema-semantics lesson of the chapter.","section":"3.5.1"}],"minor_comments":[{"comment":"The statement that the ICD-9 coding system 'consists of a five-digit structure' is an oversimplification; ICD-9-CM codes have 3 to 5 digits, and the text should acknowledge this variability.","section":"2.3.1"},{"comment":"The citation 'cite [Moody, 2009]' appears as a raw LaTeX command; it should be formatted as a proper citation.","section":"3.2"},{"comment":"The figure contains typos: 'HIPPA' should be 'HIPAA' and 'MyHeathED' should be 'MyHealthED'.","section":"Figure 3"},{"comment":"The reference list contains 'V oigt' (with a space) for 'Voigt'; this should be corrected.","section":"References"},{"comment":"The phrase 'The MIMIC-III is a relational database comprises 26 tables' is ungrammatical; it should be 'The MIMIC-III relational database comprises 26 tables'.","section":"3.3"},{"comment":"The table name 'Patients' is sometimes written as 'Patient' (e.g., 'the Patient and Icustays tables'); for consistency, use 'patients' throughout.","section":"3.4"}],"recommendation":"major_revision","confidential_remarks":"As an expository chapter, the manuscript's contribution is pedagogical rather than technical. The internal statistical inconsistency in Section 3.3, and the related age-filter inconsistency in Section 3.5.1, are precisely the kind of identifier-conflation errors the chapter warns readers about; they should be corrected and all quoted MIMIC-III summary statistics should be re-verified against the v1.4 documentation before acceptance. No concerns about citation practices or novelty are noted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a tutorial/review chapter on MIMIC-III and EHRs, not a research contribution. It does a decent job for newcomers: clear walkthroughs of schema, SQL queries, ICD coding, and a couple of case studies. The main thing to know: there is a factual error in the numbers that is exactly the kind of error the chapter warns readers against, so it needs fixing before I'd point a student at it.\n\nWhat's new: nothing. The authors acknowledge building on a Coursera specialization. The value is pedagogical. The SQL walkthroughs and entity-relationship diagrams are reimplementations of known queries, not new methods. That's fine for a book chapter, but it shouldn't be evaluated as a research paper.\n\nWhat it does well: the exposition of MIMIC-III's four table groups (patient tracking, ICU, hospital, dictionary) is accurate and easy to follow. The distinction between subject_id, hadm_id, and icustay_id is explained carefully, and the example queries for vital signs, lab events, and mortality are helpful. The section on clinical concepts (APS, SAPS, SOFA, etc.) is a reasonable summary. The catheterization study walkthrough gives a concrete sense of how a real MIMIC-based propensity-score analysis is built.\n\nSoft spots: the stress-test is right. Section 3.3 states the database 'contains over 53,000 distinct hospital admissions for patients aged 60 years and above, admitted to critical care between 2001 and 2012.' The chapter's own figures report 38,597 patients, 49,785 hospital admissions, and 53,423 ICU stays. So 53,000 is really the ICU-stay count, not hospital admissions, and there is no age-60+ restriction in that published total. That is precisely the class of error the chapter tells readers to avoid when choosing identifiers. It's not a cosmetic typo; it undermines trust in the case-study numbers. There's also an overstatement in Section 3.2 that MIMIC-III is 'the only freely available database of its kind without major usage restrictions'—that's false; eICU and AmsterdamUMCdb come to mind. The ICD-9/10 sections are simplified but acceptable for orientation.\n\nWho it's for: instructors and students who want a structured first pass at MIMIC-III. The internal inconsistency should be fixed before it's used as a primary teaching text.\n\nRecommendation: worth sending to peer review as a review/tutorial chapter, but only after the authors correct the counts and qualify the uniqueness claim. I would not cite it in my own research.","headline":"Useful MIMIC-III tutorial for newcomers, but the central counting error needs fixing before I'd trust it with students.","tokens_in":15994,"tokens_out":2913,"would_cite":false,"duration_ms":28204,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This chapter argues that MIMIC-III made critical-care data broadly accessible, and that the database's identifier structure means query choices materially change research results.","keywords":["EHR","ICD","MIMIC-III","descriptive statistics","predictive analytics","digital twins","relational database","critical care"],"falsifier":"Re-run the chapter's counting queries on the same MIMIC-III release by selecting distinct values from `subject_id`, `hadm_id`, and `icustay_id` respectively; if the three counts match, the chapter's central warning about query-sensitive results is empty rather than instructive, and if the age-filtered admission count in Section 3.3 exceeds the total in Figure 9, that specific statistic is false.","tokens_in":15096,"feed_emoji":"🏥","tokens_out":10350,"duration_ms":100501,"temperature":0.7,"pith_summary":"Electronic health records moved from paper to searchable databases, and this chapter argues that the MIMIC-III intensive-care database is the clearest proof of what that shift makes possible: a rich, real-world clinical dataset any qualified researcher can use. The chapter's central practical claim is that MIMIC-III is not one flat table but a relational database in which patient identifiers, hospital admission identifiers, and ICU-stay identifiers each answer a different question, so different querying approaches produce subtly different patient counts and outcome estimates. It demonstrates this with worked SQL queries for descriptive statistics, mortality, length of stay, vital signs, and ICD-9 coding, and uses a catheterization study to show how careful cohort selection changes conclusions. The importance is that if the authors are right, schema literacy is a prerequisite for trustworthy EHR research and for the eventual move toward predictive analytics and digital twins.","feed_headline":"How you query MIMIC-III changes the result","feed_subtitle":"The same critical-care database gives different patient counts depending on which identifier you count.","key_machinery":"The carrying mechanism is the MIMIC-III relational schema, centered on three identifiers: `subject_id` for a unique patient, `hadm_id` for a unique hospital admission, and `icustay_id` for a unique ICU stay. The chapter's argument is that these identifiers sit at different levels of granularity—one patient can have several admissions, and one admission can have several ICU stays—so counts and outcome measures computed from one identifier do not automatically transfer to another. Around this core sit the `patients`, `admissions`, and `icustays` tables, with dictionary tables that decode chart, input, output, and lab events, and with the ICD-9 code tables that attach diagnoses to admissions.","core_discovery":"The central claim is that MIMIC-III has opened up access to high-quality critical-care data, and that this access comes with a catch: the database's architecture determines the answer. Counting distinct patients, distinct hospital admissions, and distinct ICU stays yields different totals, and mortality estimates differ depending on whether death is measured in the ICU, in the hospital, or over a longer window after admission. The chapter shows these differences with reproducible queries and charts, presenting the MIMIC-III schema as the key instrument that links raw clinical events to research-grade conclusions. Along the way it treats ICD-9 codes as the clinical vocabulary that lets researchers group patients by disease, and it presents the progression from descriptive statistics through predictive models to digital twins as the direction in which EHR research is moving.","pith_inferences":["The same identifier-hierarchy problem is likely present in other large EHR databases, so the chapter's tutorial could be read as a general checklist for schema-aware reporting rather than a MIMIC-III-specific guide.","If published MIMIC-III studies rarely specify their counting semantics, a portion of apparent disagreements in the literature may be artifacts of query choice rather than genuine clinical differences.","A simple testable extension would be a benchmark task where researchers are given the chapter's clinical questions and asked to report their query definitions; the spread of answers would quantify how much architecture literacy matters.","The chapter's own statistical inconsistency (Section 3.3 versus Figures 9 and 10) illustrates the same point it argues: numbers from EHR databases need to be traced back to the exact query that produced them."],"forward_implications":["Researchers should report which identifier they counted and which mortality window they used whenever they publish MIMIC-III results, because the same database can support multiple defensible answers.","Reproducible cohort definitions, like the catheterization study's stepwise SQL pipeline, become as important as statistical models for making EHR research trustworthy.","Mortality and length-of-stay models built on MIMIC-III should be evaluated against the specific outcome definition they target, since ICU, hospital, and longer-term mortality behave differently.","The path to predictive analytics and digital twins in healthcare depends on solving the data-linkage and schema-understanding problems demonstrated here, not just on adding more data."],"supporting_citations":[{"why":"Describes the MIMIC-III database itself, the dataset whose architecture and query semantics the chapter analyzes.","marker":"[Johnson et al., 2016]"},{"why":"Provides the MIMIC code repository that makes the chapter's reproducible query approach possible.","marker":"[Johnson et al., 2018]"},{"why":"The clinical catheterization-and-mortality study that the chapter re-implements as its running example of cohort selection.","marker":"[Hsu et al., 2015]"},{"why":"Supports the chapter's claim that designing a database schema for complex physiological and clinical data is a significant challenge.","marker":"[Moody, 2009]"},{"why":"Documents the earlier MIMIC iterations and the public-access critical-care data model that MIMIC-III extends.","marker":"[Saeed et al., 2002, 2011]"},{"why":"Supports the claim that MIMIC-type big data resources enable predictive modeling and outcome research in critical care.","marker":"[Bulgarelli et al., 2020]"}],"fun_headline_variants":["Query MIMIC-III differently and mortality changes","The same MIMIC-III data yields different patient counts","How you query MIMIC-III shapes the clinical answer","MIMIC-III: Database structure dictates your result","Critical care numbers shift with MIMIC-III query choices"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The chapter's usefulness rests on the factual accuracy of its MIMIC-III statistics, and that accuracy is internally inconsistent: Section 3.3 reports over 53,000 hospital admissions for patients aged 60 and above, while Figures 9 and 10 report 49,785 hospital admissions and 53,423 ICU stays.","fun_headline_variants_meta":{"raw":{"variants":["Query MIMIC-III differently and mortality changes","The same MIMIC-III data yields different patient counts","How you query MIMIC-III shapes the clinical answer","MIMIC-III: Database structure dictates your result","Critical care numbers shift with MIMIC-III query choices"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000224,"raw_usage":{"total_tokens":1462,"prompt_tokens":951,"completion_tokens":511,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":567,"completion_tokens_details":{"reasoning_tokens":433}},"tokens_in":567,"tokens_out":511,"duration_ms":5378,"temperature":1.0,"reasoning_tokens":433,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T19:47:43.600361+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the chapter's counting queries on the same MIMIC-III release by selecting distinct values from `subject_id`, `hadm_id`, and `icustay_id` respectively; if the three counts match, the chapter's central warning about query-sensitive results is empty rather than instructive, and if the age-filtered admission count in Section 3.3 exceeds the total in Figure 9, that specific statistic is false.","supporting_citations":[{"cited_title":"The mimic code repository: enabling reproducibility in critical care research","cited_arxiv_id":null,"evidence_quote":"Provides the MIMIC code repository that makes the chapter's reproducible query approach possible."},{"cited_title":"The association between indwelling arterial catheters and mortality in hemodynamically stable patients with respiratory failure: a propensity score analysis","cited_arxiv_id":null,"evidence_quote":"The clinical catheterization-and-mortality study that the chapter re-implements as its running example of cohort selection."},{"cited_title":"Physionet: Research resource for complex physiologic signals","cited_arxiv_id":null,"evidence_quote":"Supports the chapter's claim that designing a database schema for complex physiological and clinical data is a significant challenge."},{"cited_title":"Mimic ii: a massive temporal icu patient database to support research in intelligent patient monitoring","cited_arxiv_id":null,"evidence_quote":"Documents the earlier MIMIC iterations and the public-access critical-care data model that MIMIC-III extends."},{"cited_title":"big data","cited_arxiv_id":null,"evidence_quote":"Supports the claim that MIMIC-type big data resources enable predictive modeling and outcome research in critical care."}],"review_version":1}