Inputs and outputs
Inputs and outputs are extremely useful when studying intensive care unit patients. Inputs are any fluids which have been administered to the patient: such as oral or tube feedings or intravenous solutions containing medications. Outputs are fluids which have either been excreted by the patient, such as urine output, or extracted from the patient, for example through a drain. These data were the most complicated to handle technically in the MIMIC-III data.
The MIMIC-III database contains information from two distinct critical care information systems: Philips CareVue and iMDSoft Metavision. These two databases store data in different ways. In the descriptions in this document, data will be referred to as being sourced either from “CareVue” or “Metavision” to differentiate between the different systems.
INPUTEVENTS_CV, INPUTEVENTS_MV, OUTPUTEVENTS
Inputs exist in two separate tables: INPUTEVENTS_CV and INPUTEVENTS_MV. INPUTEVENTS_CV contains CareVue inputs, while INPUTEVENTS_MV contains Metavision inputs. Results from these tables can be unioned as the same patient stay never occurs in both tables. Concretely, a unique
ICUSTAY_ID only occurs in one out of the two tables. However, if a patient has two ICU stays, one in CareVue and one in Metavision, then they could appear in both tables. As a result, a unique
SUBJECT_ID could occur in both tables.
All outputs for both Metavision and CareVue patients have been merged into the OUTPUTEVENTS table.
Outputs are recorded in a reasonably consistent manner in both the CareVue and Metavision data. The time at which the output is measured is recorded in the
CHARTTIME column. There is no start time recorded with outputs -
CHARTTIME corresponds to the time that the volume had been output by. The volume of output is recorded in the
VALUE column, and the unit of measurement is provided in the
VALUEUOM column (usually milliliters, or mL).
It is usually reasonable to assume that any output recorded is for the interval between the current
CHARTTIME and the previous
CHARTTIME for the same item.
Inputs are handled differently by CareVue and Metavision. For CareVue data, only the
CHARTTIME is available. Second, the
AMOUNT columns are asynchronous, and originally stored in different tables. Volumes of input (e.g. 50 mL of normal saline) would be stored in one table (the original IOEVENTS), and would usually be recorded every hour (though sometimes the period was longer). Conversely, the
RATE of the drug would be stored separately (in MEDEVENTS), and only updated when a change or verification of the rate was made by clinical staff. As a result, the raw data looked something similar to:
Here, the volume is recorded only every hour, and no start time is available. However, it’s reasonable to assume that the volume measurement corresponds to an hour. Next, we can see that the rate was titrated to 0.5, and for the period between 11:00 to 12:00 there was half an hour of delivery at 1 mL/min, and half an hour of delivery at 0.5 mL/min, resulting in a total volume of 45 mL delivered for the past hour.
Summing up, for CareVue data, the rate and volume will be asynchronous, and only the
CHARTTIME will be available. For rates, the
CHARTTIME will correspond to a start time (when the drug was set to that rate). For volumes, the
CHARTTIME will correspond to an end time.
For Metavision data, there is no concept of a volume in the database: only a
RATE. All inputs are recorded with a
STARTTIME and an
ENDTIME. As a result, the volumes in the database for Metavision patients are derived from the rates. Furthermore, exact start and stop times for the drugs are easily deducible.
Details of the merging process
The difficulty in merging the databases arose due primarily to two factors: the lower resolution of information archiving in the CareVue system, and the different definition of an ‘order’ in the databases.
The aim of this section is to provide all the detail into how these data were merged: this information is not necessary to understand for the purposes of using the database, but will provide insight into the format of the IOEVENTS table. We welcome suggestions from the community on improving the format and usability of the table.
The CareVue system stored input/output (IO) data across five tables: IOEVENTS, MEDEVENTS, ADDITIVES, SOLUTIONS and DELIVERIES. Each time a new order for a drug was recorded in the database, the ADDITIVES, SOLUTIONS and DELIVERIES tables would be populated with information regarding the order. The data archival format is best described with an example.
Patient A has been recently admitted to the ICU and is to be administered noradrenaline to restore their blood pressure to a value of at least 60 mmHg. The route of administration is intravenous, i.e. the drug is to be pumped into the patient’s blood stream directly through a line inserted in a vein. The nurse would prepare a solution of 250 mL sodium chloride (NaCl) to contain the drug. The SOLUTIONS table would consequently contain an entry of 250 mL NaCl for patient A at the time the nurse prepared the solution. The nurse would then mix in the drug into the solution: in this example let’s say 8 mg of noradrenaline. The 8 mg of noradrenaline would be recorded in the ADDITIVES table, and an identifier would be recorded which linked the 8 mg of noradrenaline to the 250 mL solution of NaCl. Finally, the nurse would optionally set an initial delivery rate and route: in this case it could be 10 mL/hr intravenously. These would populate the
ROUTE columns in the DELIVERIES table. Note that DELIVERIES data is less frequently present and was not consistently recorded.
Now that the solution has been prepared, the nurse can begin administering the drug to the patient. The time is now 18:20. An initial entry of 0 mL is recorded in the IOEVENTS table at 18:00 (this usually occurs, but it has not been verified that this always occurs). The nurse begins at a rate of 1 mcg/kg/min. The MEDEVENTS contains an entry at a
CHARTTIME of 18:20 for a rate of 1 mcg/kg/min. Five minutes later, at 18:25, the nurse notes that the blood pressure is still lower than the desired 60 mmHg and increases the dose. The nurse raises the dose to 2 mcg/kg/min, and MEDEVENTS records the new dose of 2 mcg/kg/min at 18:25. The nurse checks again and notes that the blood pressure has reached the target value of 60 mmHg and ceases titration of the drug. At 19:00, the volume of drug administered to the patient is recorded. If the patient weighed 100 kg, then the amount of drug administered would be equal to:
5 min * 1 mcg/kg/min * 100 kg + 25 min * 2 mcg/kg/min * 100 kg = 500 mcg + 5000 mcg = 5500 mcg = 5.5 mg
The amount of solution administered would be equal to:
( 5 min * 1 mcg/kg/min * 100 kg + 25 min * 2 mcg/kg/min * 100 kg ) * 250 mL / 8 mg = 5.5 mg * 250 mL / 8 mg = 171.875 mL
Consequently, IOEVENTS would record 171.875 mL at 19:00. If the patient continued at the same rate for this drug, then IOEVENTS would record (60 * 2 * 100 * 250 / 8000) = 375 mL at 20:00. Note that:
- Unless the rate is updated, no new entry in MEDEVENTS exists
- IOEVENTS always records a value on an hourly basis, regardless of how long the IO event has been present
- 375 mL is larger than the original 250 mL bag: somewhere during administration the nurse would need to replace the empty bag with a new solution of the same formulation
Note that the changing of the bag is sometimes, but not always, recorded in the ADDITIVES/SOLUTIONS/DELIVERIES table as a new order. It usually occurs if the drug has been discontinued, and then re-prescribed later in the patient’s stay.
Metavision records IO data using two tables: RANGESIGNALS and ORDERENTRY. These tables do not appear in MIMIC-III as they have been merged to form the INPUTEVENTS_MV table. RANGESIGNALS contains recorded data elements which last for a fixed period of time. Furthermore, the RANGESIGNALS table recorded information for each component of the drug separately. For example, for a noradrenaline administration there would be two components: a main order component (noradrenaline) and a solution component (NaCl). The
ENDTIME of RANGESIGNALS indicated when the drug started and finished. Any change in the drug rate would result in the current infusion ending, and a new
STARTTIME being created.
Let’s return to our example above of the patient being given noreadrenaline. The
STARTTIME for the solution (NaCl) and the drug (noradrenaline) would be 18:20. The rate of the drug would be listed as 1 mcg/kg/min, and the rate of the solution would be listed as 10 mL/hr. The nurse, as before, decides to increase the drug rate at 18:25 to 2 mcg/kg/min. As a result, the
ENDTIME for the two rows corresponding to the solution (NaCl and noreadrenaline) is set to 18:25. Two new rows are generated with a
STARTTIME of 18:25. These two new rows would continue until either (i) the drug rate was changed or (ii) the drug was delivery was discontinued. The
ORDERID column would be identical for each instantiation of NaCl and noradrenaline which corresponded to the same solution/rate. That is, for the infusion given between 18:20 and 18:25, both NaCl and noreadrenaline would have the same
LINKORDERID would further link the drug across all administrations, even when the rate is changed. The following table demonstrates these concepts:
ORDERID links items occurring at the same time which correspond to the same solution, while
LINKORDERID links all these solutions together across time. Note also that
LINKORDERID is equal to the first
ORDERID which occurs for the solution, as above.