1.4.2 What is a patient record?

Since a patient may have been admitted several times during the period in which our data were collected, it is important to understand exactly how to identify patients and their individual patient stays.

There are essentially four identifiers for data associated with any given patient:

• Subject ID (Subject_ID) - an integer number identifying a particular patient. This can be thought of as a substitute for a unique medical record number. In the flat file data posted on PhysioNet, the number representing the Subject_ID is left padded with zeros to five digits and preceded by the letter s. In the relational database, the Subject_ID is an integer and therefore has no preceding letter.
• Hospital admission ID (Hadm_ID) - an integer number identifying a particular admission to the hospital. Each patient may have many Hadm_ID's associated with their unique Subject_ID.
• ICU stay ID (ICUstay_ID) - an integer number identifying an ICU stay. An ICU stay, refers to the period of time when the patient is cared for continuosly in an Intensive Care Unit. Each patient may have one or more ICU stays associated. An ICU stay is considered to be continuous if any set of ICU events (such as bed transfers or changes in type of service) belonging to one Subject_ID are less than 24 hours apart. Longer breaks in the patient's stay automatically cause a new ICUstay_ID to be assigned.
• Case ID (Case_ID) - This is a five digit number preceded by the letter a (for adults) or n (for neonates). This ID indicates a set of waveforms associated with a given patient. For various reasons (described in section: 1.4.3 below), there may be multiple case IDs associated with a given patient.

Figure 1.4 illustrates the connection between the above labels for a given individual, with each hospital stay surrounded by a red box. Time progresses from left to right, and the type/quantity of information is represented by the vertical axis. The first piece of information concerning a patient is recorded at time A and the patient is then admitted to the first ICU (ICU1) at time B. The first ICU data point is collected about the patient at time C. Notice also that the amount of information reduces part way through the ICU stay (indicated by the contraction of the vertical width of the blue box). This is common in many patients, where less signals are monitored as a patient improves. Note that at time D the waveforms and trends end. This may be because the patient was disconnected from the bedside monitors, prior to discharge, or because there was some issue that led to an interruption in data collection. At time E (which is more than 24 hours after the discharge from ICU1, the patient is readmitted to the ICU (although this time it is a different ICU (ICU2). Note that this is not always the case, a patient may be discharged and re-admitted to the same or a different ICU. If the time is less than 24 hours, then the patient is considered to be still in the ICU and no new ICU stay is created, even if they are transferred to a new ICU (e.g. ICU3). After discharge (at time H) and readmittance (at time I) longer than 24 hours, a new ICU stay (3) is created.

Note that a patient may move between ICUs during any given admission. If the move is longer than 24 hours, we define it to be a new admission. Note also that the amount of data varies during and between ICU stays and that data are often missing - see section 1.6.

Note also that for the preliminary release, any patients that have stays which overlap their 90$^{th}$ birthday have been removed from the database. Patients used in the test set for the PhysioNet / Computers in Cardiology Competition 2009 have also been removed in this first release.