The MIMIC Code Repository: enabling reproducibility in critical care research

Johnson, Alistair E. W.; Stone, David J.; Celi, Leo Anthony; Pollard, Tom

doi:10.1093/jamia/ocx084

Cited by 278 publications

(201 citation statements)

References 31 publications

Supporting

Mentioning

199

Contrasting

Order By: Relevance

“…Structured query language and codes from the MIMIC Code Repository (https://github.com/MIT-LCP/mimiccode) 26 were used to extract data from the database, including age, sex, ethnicity, type of admission, Simplified Acute Physiology Score II (SAPS II) 27 on admission, mechanical ventilation on first day, renal replacement therapy on first day, Elixhauser Comorbidity Index (SID30), 28 and specific comorbidities. For patients who were older than 89 years, the database shifted the date of birth to exactly 300 years before to obscure their ages, and these records were corrected (by minus 300 and plus 89) before analysis.…”

Section: Variables Extractionmentioning

confidence: 99%

<p>Association Between Blood Glucose Within 24 Hours After Intensive Care Unit Admission and Prognosis: A Retrospective Cohort Study</p>

Chen

et al. 2020

DMSO

View full text Add to dashboard Cite

The aim of this study was to evaluate the association between blood glucose within 24 hours after intensive care unit (ICU) admission and prognosis. Patients and Methods: A retrospective cohort study was conducted using data from a large critical care database. Patients who had a length of ICU stay ≥24 hours and at least two blood glucose records within 24 hours after ICU admission were included and hospital mortality was chosen as the primary outcome. The average, minimum, and maximum blood glucose within 24 hours after ICU admission were a priori selected as exposures and associations between each exposure and outcomes were assessed after adjusted for potential confounders. Results: A total of 14,237 patients were included finally with an average age of 62.9±17.7 years and a mean SAPS II on admission of 34 (26-44). Among the study population, 20.2% (2872/14,237) had uncomplicated diabetes, and 6.7% (953/14,237) had complicated diabetes. Lowest hospital mortality rate was observed in the stratum with an average blood glucose ranged 110-140 mg/dL, a minimum blood glucose ranged 80-110 mg/dL, and a maximum blood glucose ranged 110-140 mg/dL. After adjusted for confounders including age, sex, disease severity scores and comorbidities, an average blood glucose ranged 110-140 mg/dL, a minimum blood glucose ranged 80-110 mg/dL, and a maximum blood glucose ranged 110-140 mg/dL were associated with the lowest risk of hospital mortality. Consistent results were found among patients without diabetes in the subgroup analyses stratified by diabetes. Conclusion: A range of 110-140 mg/dL for average and maximum blood glucose and a range of 80-110 mg/dL for minimum blood glucose within 24 hours after ICU admission predicted better prognosis especially among patients without diabetes.

show abstract

Section: Variables Extractionmentioning

confidence: 99%

<p>Association Between Blood Glucose Within 24 Hours After Intensive Care Unit Admission and Prognosis: A Retrospective Cohort Study</p>

Chen

et al. 2020

DMSO

View full text Add to dashboard Cite

show abstract

“…Septic patients in our study were identi ed from the Medical Information Mart for Intensive Care (MIMIC-III) database. This database included information on 53,423 hospital admissions of patients (16 years or older) in the ICU of Beth Israel Deaconess Medical Center from June 1, 2001 to October 31, 2012 in Boston, Massachusetts [12]. Since the database was provided anonymously and publicly by a third party (the Massachusetts Institute of Technology (MIT) Laboratory for Computational Physiology) with prior approval from the Institutional Review Board (IRB) [13], the requirements for IRB and informed consent approval for the study were waived.…”

Section: Methodsmentioning

confidence: 99%

Time to Lactate Measurement Affects Efficacy of Initial Lactate on Mortality in critically ill patients with sepsis.

Luo¹,

Zheng²,

Rao³

et al. 2020

Preprint

View full text Add to dashboard Cite

Background: Evidence regarding the effect of time to lactate measurement on the relationship between the initial lactate level and mortality is limited. We aimed to investigate the relationships between time to lactate measurement, initial lactate level, and in-hospital mortality in critically ill patients with sepsis.Methods and Results: Of the 14339 eligible adult patients with recognized sepsis upon admission to the ICU based on the MIMIC-III database, the median value of initial lactate was 1.70 mmol/L (interquartile range [IQR] 1.20-2.80), and its detection time was 3.50 hours ([IQR] 1.31-10.24). The results of fully adjusted multivariate analyses demonstrated that lactate was positively associated with in-hospital mortality (odds ratio: 1.126, 95% confidence interval: 1.090 to 1.163, P<0.001), and there was an increase in the odds of death with hourly delays in lactate measurement (OR: 1.006, 95% CI: 1.004 to 1.008, P<0.001). In stratified analyses, delays in lactate measurement significantly interfered with the impact of increased lactate level on mortality (P-value for interaction<0.001). The hospital mortality rate substantially increased by 43.5% for each unit increase in lactate when measurement was delayed by 24 hours (OR: 1.435, 95% CI: 1.260 to 1.635, P<0.001).Discussion: The association of initial lactate with in-hospital mortality is likely to vary with delays in detection time (grouping based on the “1-hour bundle”) in critically ill patients with recognized sepsis upon admission to the ICU.

show abstract

“…The fact that intubated patients have been assigned the lowest possible GCS in the MIMIC-II dataset has largely been ignored in the literature. It was briefly mentioned by the PhysioNet team in the calculation of the sequential organ failure assessment (previously known as sepsis-related organ failure assessment; SOFA) score in the MIMIC-III dataset (16). Table 1, bottom, shows the top 5% of the rules where the experts' ranking is lower than the empirical ranking.…”

Section: Delta Rank Helps Discover Hidden Confoundersmentioning

confidence: 99%

Expert-augmented machine learning

Gennatas

Friedman

Ungar

et al. 2020

Proc. Natl. Acad. Sci. U.S.A.

View full text Add to dashboard Cite

Machine learning is proving invaluable across disciplines. However, its success is often limited by the quality and quantity of available data, while its adoption is limited by the level of trust afforded by given models. Human vs. machine performance is commonly compared empirically to decide whether a certain task should be performed by a computer or an expert. In reality, the optimal learning strategy may involve combining the complementary strengths of humans and machines. Here, we present expert-augmented machine learning (EAML), an automated method that guides the extraction of expert knowledge and its integration into machine-learned models. We used a large dataset of intensive-care patient data to derive 126 decision rules that predict hospital mortality. Using an online platform, we asked 15 clinicians to assess the relative risk of the subpopulation defined by each rule compared to the total sample. We compared the clinician-assessed risk to the empirical risk and found that, while clinicians agreed with the data in most cases, there were notable exceptions where they overestimated or underestimated the true risk. Studying the rules with greatest disagreement, we identified problems with the training data, including one miscoded variable and one hidden confounder. Filtering the rules based on the extent of disagreement between clinician-assessed risk and empirical risk, we improved performance on out-of-sample data and were able to train with less data. EAML provides a platform for automated creation of problem-specific priors, which help build robust and dependable machine-learning models in critical applications.

show abstract

The MIMIC Code Repository: enabling reproducibility in critical care research

Abstract: By providing open source code alongside the freely accessible MIMIC-III database, we enable end-to-end reproducible analysis of electronic health records.

Cited by 278 publications

References 31 publications

<p>Association Between Blood Glucose Within 24 Hours After Intensive Care Unit Admission and Prognosis: A Retrospective Cohort Study</p>

<p>Association Between Blood Glucose Within 24 Hours After Intensive Care Unit Admission and Prognosis: A Retrospective Cohort Study</p>

Time to Lactate Measurement Affects Efficacy of Initial Lactate on Mortality in critically ill patients with sepsis.

Expert-augmented machine learning

Contact Info

Product

Resources

About