A Temporal Dataset for Respiratory Support in Critically Ill Patients

Abstract

We present a temporal benchmark dataset for clinical respiratory intervention tasks in intensive care unit (ICU) patients, derived from the MIMIC v2.2 dataset. The data consists of 50,920 adult ICU patients and includes 90-day hourly ventilation data, laboratory results, vital signs, and details of treatment interventions such as respiratory support interventions, vasopressor administration, and continuous renal replacement therapy. Additionally, it encompasses static variables such as ICU length of stay, discharge outcome, and death outcome. While there are existing open ICU datasets, the limited availability of benchmark datasets designed for clinical treatments in respiratory support motivated the development of the data. The data facilitates various tasks including time-series analysis, survival analysis, sequential decision-making tasks, and reinforcement learning.

Background

Critically-ill patients often find themselves in the intensive care unit (ICU) seeking specialized support for respiratory distress  [1]. Despite advances in supportive treatments, the in-hospital mortality rate remains 40% for conditions such as acute lung injury and acute respiratory distress syndrome. Managing respiratory distress involves intricate treatment measures, including invasive mechanical ventilation, non-invasive mechanical ventilation, and high-flow nasal cannula. However, existing recommendations and outcomes, especially regarding intubation and weaning procedures for ICU patients, remain controversial and poorly understood [2].

In complex ICU environments, well-curated observational data is crucial for evidence-based decision-making, especially given the challenges of conducting real-world trials for respiratory treatments that evolve over time. Observational health data derived from Electronic Health Records (EHRs) presents a valuable resource. While publicly available de-identified EHR ICU datasets exist, such as HiRID, Salzburg Intensive Care database, AmsterdamUMCdb, and the Medical Information Mart for Intensive Care (MIMIC) [3], existing benchmarks primarily focus on conventional clinical prediction tasks like mortality prediction and length-of-stay estimation [4-6].

To address this gap, we present a temporal benchmark dataset derived from MIMIC-IV (v2.2) for evaluating respiratory intervention tasks. The hourly dataset is enriched with ventilation data and a wide range of other covariates, including demographics, lab results, measurements, illness severity scores, treatment interventions, and outcome variables. It covers 50,920 patients admitted to the ICU, with records collected over 90 days. This dataset can help address weaning delays and failures, optimize strategies for respiratory support, identify efficiencies in clinical practices, provide decision support to attending physicians regarding intubation decisions in the ICU, and facilitate time-series and reinforcement learning applications. It can also be linked to social determinants of health to better understand societal implications and biases in complex healthcare settings [7-11] . 

Methods

Data Description

The dataset presented comprises 50,920 distinct adult patients admitted to the ICU of Beth Israel Deaconess Medical Center (Boston, MA, USA) between 2008 and 2019. Static, time-varying, and outcome variables have been extracted from MIMIC-IV and organized into an hourly materialized view. The content for each patient is stored in a *.csv format, named after the patient's unique identifier (subject ID). The directory where the data is stored, is organized into multiple folders, each named according to the first three digits of the subject IDs. Within each folder, CSV files are stored for all subjects whose IDs begin with the corresponding three digits of the folder name.

Patient level static variables

Static parameters extracted for patients, as outlined in Table 1, encompass demographic variables, comorbidity scores that evaluate neurological function (including the Glasgow Coma Scale and its three components: eye opening, verbal and motor responses), along with an assessment of patient organ dysfunction (maximum Sequential Organ Failure Assessment score) performed 24 hours after ICU admission.

Measurements/observations

The time-varying measurements in the data encompass ventilation settings, laboratory results, and vital signs. Ventilation settings and vital signs are extracted from the MIMIC chartevents table, while labs data are obtained from the MIMIC labevents table, each identified by their respective ItemIDs. If there were multiple ItemIDs representing the same measurement with different units, we standardize the units before incorporating them into our data. To handle multiple values within a single hour for a subject, the results are aggregated by computing the median, given that the median exhibits reduced sensitivity to noisy data. The labs are sourced from arterial blood gas (ABG) specimens, as arterial blood measurements are deemed to have greater clinical relevance and precision when evaluating parameters such as respiratory function, acid-base balance, and oxygenation status. Two parameters derived from ventilation settings are also presented: set_pc_draeger (set pressure for pressure-controlled ventilation from the Draeger ventilator) and set_pc (set pressure for pressure-controlled ventilation). Set_pc_draeger is calculated as the difference between the inspiratory pressure from the Draeger ventilator (pinsp_draeger) and the set peak inspiratory pressure (ppeak). Based on clinical knowledge, set_pc is populated with pcv_level (pressure controlled ventilation level) if present, pinsp_hamilton (inspiratory pressure from Hamilton ventilator) if pcv_level is absent, and set_pc_draeger (inspiratory pressure from Draeger ventilator) if both are absent. All variables related to ventilation parameters, vital signs, and labs, and their corresponding descriptions, are described in Table 2 within their respective sections.

¹ "set" in ventilation settings refers to values set by healthcare professionals on the ventilator to suit the patients’ respiratory needs.

² "_" refers to intermediate variables.

Treatment interventions

Respiratory support interventions

Three respiratory support methods, namely invasive ventilation (INV), non-invasive ventilation (NIV), and high-flow nasal cannula (HFNC), are presented as binary indicators per hour. The curation of these respiratory support variables is verified by clinical experts to ensure accuracy and reliability. 

In  MIMIC, the procedureevents table identifies patients on INV or NIV during their ICU stay, while the chartevents table identifies patients on HFNC. INV and NIV in MIMIC have documented start and end times recorded by respiratory therapists, however, HFNC lacks a corresponding time interval; having only the time at which the measurement was observed.

Therefore, we pre-process the data to establish a start time and end time for each HFNC event per ICU stay. In addition, multiple HFNC events could occur during a single ICU stay. Therefore, if the time gap between two consecutive HFNC events exceeded 24 hours, we treat them as separate events. For each HFNC event, the minimum and maximum time at which the HFNC is applied is used to obtain the HFNC start time and the end time. HFNC events with identical start and end times are excluded.

In cases of overlapping mutually exclusive treatments, where patients were recorded to be on both non-invasive and invasive ventilation simultaneously, we prioritize the most invasive treatment strategy (INV > NIV > HFNC). This occurs due to the complexities involved in transitioning between ventilation therapies within the ICU, which often includes a series of procedures during the transition period. Furthermore, for short intervals (less than 6 hours) recorded between two different treatments, we attribute the gap to the less invasive treatment. This allows us to handle situations where the precise timing of treatments is unclear. 

Additional interventions

Additional binary indicators for interventions include vasopressor administration and continuous renal replacement therapy. Vasopressors are extracted from the MIMIC inputevents table and matched to the corresponding hour using their respective start and end times. A patient is classified as being on vasopressors if they received norepinephrine, epinephrine, dopamine, phenylephrine, or vasopressin. Information regarding continuous renal replacement therapy (CRRT) is extracted from the MIMIC chartevents table. Patients are identified as being on CRRT if they had a positive value for blood flow rate or fluid removal during dialysis.  A summary of all treatment interventions is presented in Table 3. 

Outcome Variables

The majority of the outcome variables are recorded as binary markers at each hour, with one denoting the occurrence of the event. These include discharge outcome, ICU out-time outcome, death outcome, sepsis. 

Discharge outcome and ICU out-time outcome indicate if a patient was discharged from the hospital or ICU respectively. The death outcome variable denotes whether a patient died at a specific hour. The date of death in MIMIC was derived from hospital and state records. In cases where both data sources are available, in-hospital mortality is preferentially used over state-linked data. The state-derived date of death included only the date component, so a default time of midnight is used when converting the date to a timestamp.

The data also includes a sepsis outcome variable that identifies whether a patient is septic according to the Sepsis-3 diagnostic criteria. Additionally, it contains the length of stay variable, which indicates the duration of a patient's ICU stay in fractional days.  A summary the outcome variables is presented in Table 4.

Usage Notes

The dataset can be utilized in various studies focusing on respiratory support and ICU practices, addressing issues such as weaning delays and optimizing intubation decisions. This dataset offers considerable reuse potential in multiple areas. Researchers can use it to address weaning delays and failures, optimize respiratory support strategies, identify efficiencies in clinical practices, and provide decision support to attending physicians regarding intubation decisions in the ICU. Additionally, it facilitates time-series analysis and reinforcement learning approaches, aiding in predictive modeling and decision-making processes.

One limitation to consider is the generalizability of the dataset, as the dataset may not fully represent all ICU settings, particularly those with differing protocols or patient populations. Temporal bias is another limitation, as changes in clinical practices over time could impact the relevance of older data within the dataset.

Ethics

The dataset originates from the MIMIC-IV v2.2 database, which is a de-identified dataset accessed through the PhysioNet Credentialed Health Data Use Agreement (v1.5.0) that we have been granted permission to use.The dataset is a derivative dataset and thus no new patient data was collected. The ethics approval of the dataset follows from that of the parent MIMIC dataset.

Acknowledgements

L.C., D.M., and L.M. are supported by grant NIH-R01-EB017205 from the National Institute of Health. D.M. and L.M. are supported by the National Institute of Biomedical Imaging and Bioengineering (NIBIB) under NIH-R01-EB030362. D.M. is also supported by NIH National Library of Medicine under 75N97020C00013, and Massachusetts Life Sciences Center. H.C. is supported by the Korea Health Technology Research and Development Project through the Korea Health Industry Development Institute, funded by the Ministry of Health and Welfare, Republic of Korea (grant number: HI21C107409). The authors would like to thank Dr. Kerollos Wanis, for his valuable feedback.

Conflicts of Interest

The authors have no conflicts of interests to declare.

References

1. Doyle RL, Szaflarski N, Modin GW, Wiener-Kronish JP, Matthay MA. Identification of patients with acute lung injury. Predictors of mortality. American journal of respiratory and critical care medicine. 1995 Dec;152(6):1818-24.
2. Wanis KN, Madenci AL, Hao S, Moukheiber M, Moukheiber L, Moukheiber D, Moukheiber S, Young JG, Celi LA. Emulating target trials comparing early and delayed intubation strategies. Chest. 2023 Oct 1;164(4):885-91.
3. Johnson AE, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, Pollard TJ, Hao S, Moody B, Gow B, Lehman LW. MIMIC-IV, a freely accessible electronic health record dataset. Scientific data. 2023 Jan 3;10(1):1.
4. Harutyunyan H, Khachatrian H, Kale DC, Ver Steeg G, Galstyan A. Multitask learning and benchmarking with clinical time series data. Scientific data. 2019 Jun 17;6(1):96.
5. Purushotham, S., Meng, C., Che, Z. & Liu, Y. Benchmarking deep learning models on large healthcare datasets. J.236 biomedical informatics 83, 112–134 (2018)
6. Gupta M, Gallamoza B, Cutrona N, Dhakal P, Poulain R, Beheshti R. An extensive data processing pipeline for mimic-iv. InMachine Learning for Health 2022 Nov 22 (pp. 311-325). PMLR.
7. Moukheiber D, Mahindre S, Moukheiber L, Moukheiber M, Gao M. Looking Beyond What You See: An Empirical Analysis on Subgroup Intersectional Fairness for Multi-label Chest X-ray Classification Using Social Determinants of Racial Health Inequities. arXiv preprint arXiv:2403.18196. 2024 Mar 27.
8. Nazer LH, Zatarah R, Waldrip S, Ke JX, Moukheiber M, Khanna AK, Hicklen RS, Moukheiber L, Moukheiber D, Ma H, Mathur P. Bias in artificial intelligence algorithms and recommendations for mitigation. PLOS digital health. 2023 Jun 22;2(6):e0000278.
9. Celi LA, Cellini J, Charpignon ML, Dee EC, Dernoncourt F, Eber R, Mitchell WG, Moukheiber L, Schirmer J, Situ J, Paguio J. Sources of bias in artificial intelligence that perpetuate healthcare disparities—A global review. PLOS Digital Health. 2022 Mar 31;1(3):e0000022.
10. Nakagawa K, Moukheiber L, Celi LA, Patel M, Mahmood F, Gondim D, Hogarth M, Levenson R. AI in pathology: what could possibly go wrong?. InSeminars in Diagnostic Pathology 2023 Mar 1 (Vol. 40, No. 2, pp. 100-108). WB Saunders.
11. Charpignon ML, Byers J, Cabral S, Celi LA, Fernandes C, Gallifant J, Lough ME, Mlombwa D, Moukheiber L, Ong BA, Panitchote A. Critical bias in critical care devices. Critical Care Clinics. 2023 Oct 1;39(4):795-813.