Dark Mode
SUPPORT Critical Illness Survival Data
Public Health & Epidemiology
Tags and Keywords
Trusted By
"No reviews yet"
Free
About
Comprises records for 9,105 critically ill patients across five medical centres in the United States. The material was accessioned across two distinct periods, 1989–1991 and 1992–1994. The primary aim is to establish prognostic models for estimating patient survival rates over a 180-day period (2- and 6-month survival). The records cover nine serious illness categories, including acute respiratory failure, liver disease, coma, and various cancers. This resource provides information on demographics, disease severity, and physiological variables, addressing concerns related to patient autonomy at the end of life and supporting earlier end-of-life decision-making and planning.
Columns
The dataset contains a subset of 10 columns across 9,105 records, all of which are 100% valid:
- age: The age of the patient.
- death: A measurement related to mortality, ranging from 18 to 102, with an average of 62.7.
- sex: The gender of the patient, categorised as male (56%) or female (44%).
- hospdead: Indicates whether the patient died while hospitalised.
- slos: Short/Long classification of the length of the patient's stay.
- d.time: The time of death, measured in days, ranging from 3 to 343, with a mean of 17.9 days.
- dzgroup: The patient's disease group.
- dzclass: The specific disease class, with eight unique categories. Acute Respiratory Failure/Multiple Organ System Failure (ARF/MOSF) with Sepsis is the most common class (39%).
- num.co: The number of co-morbidities the patient has, with ARF/MOSF being the most frequent co-morbidity category (46%).
- edu: The patient's education level, ranging from 0 to 9, with an average of 1.87.
Distribution
The material is stored in a CSV file named
support2.csv, which is 3.14 MB in size. The dataset includes 9,105 records, representing individual patient cases. The available sample includes 10 of the original 47 columns. Data quality is strong, as all included fields are 100% valid, with no missing records. Although no specific preprocessing was performed, suggested default values are available for imputing high percentages of missing baseline physiological data, such as setting serum albumin to 3.5. The expected update frequency is Never.Usage
This resource is suited for developing and validating statistical prognostic models, particularly for estimating survival probability up to 180 days for seriously ill adults. It allows researchers to analyse the factors contributing to survival and compare model predictions against independent physician estimates. The data supports studies in healthcare quality improvement, end-of-life care protocols, and patient prognosis prediction using standard model selection techniques like train-test or three-way holdout splits.
Coverage
The geographic scope includes patients across five United States medical centres. The temporal scope spans two phases: Phase I patients accessioned during 1989–1991 and Phase II patients accessioned during 1992–1994. Patients were tracked for six months, and mortality information was matched against the National Death Index (NDI) through 1997. The content focuses on critically ill hospitalised patients across nine defined disease categories and includes sensitive demographic data like race, gender, income, and education level.
License
Attribution 4.0 International (CC BY 4.0)
Who Can Use It
The dataset is intended for clinicians, medical researchers, statisticians, and data scientists. It is particularly valuable for those working on mortality prediction, hospital resource optimisation, and studies focused on improving decision-making regarding end-of-life care for seriously ill patients. The material holds a maximum usability rating of 10.00.
Dataset Name Suggestions
- SUPPORT Critical Illness Survival Data
- Hospitalized Patient Prognosis and Mortality Records
- US Critically Ill Patient Survival Study 1989-1994
Attributes
Original Data Source: SUPPORT Critical Illness Survival Data
QUALITYLoading...