Compute daily census

Calculates the daily number of patients that were hospitalized at each site during the specified time period. The daily census is defined as a cross-sectional count of bed occupancy at a given time of the day (8am by default). It is calculated as the number of patients who were admitted before and discharged after that point in time (i.e., the number of patients who occupied a bed at 8am on each day). Results are returned as raw counts as well as a capacity ratio (= census/capacity). The capacity ratio indicates whether the number of hospitalized patients on a particular day was above (>1) or below (<1) the typical occupancy (or max occupancy) during the time period of interest. By default, capacity is estimated as median(census) but users can specify other measures of typical/max capacity (see capacity_func input). Note that if cohorts are filtered/grouped based on patient characteristics (e.g., diagnosis, age etc.), this indicator simply reflects a standardized measure of patient counts, rather than a measure of true capacity limitations of medical wards.

Usage

daily_census(
  cohort,
  time_period = NULL,
  scu_exclude = NULL,
  group_var = NULL,
  capacity_func = "median",
  buffer = 30,
  time_of_day = "08:00:00",
  include_zero = TRUE
)

Arguments

cohort

(data.frame or data.table) Cohort table with all relevant encounters of interest, where each row corresponds to a single encounter. Must contain the following columns:

genc_id (integer): GEMINI encounter ID
hospital_num or hospital_id (integer or character): Hospital ID
admission_date_time (character | POSIXct POSIXt): Date-time of admission in YYYY-MM-DD HH:MM format
discharge_date_time (character | POSIXct POSIXt): Date-time of discharge in YYYY-MM-DD HH:MM format

If a group_var is specified (see below), this should be included in the cohort table as well.

time_period

(character) The start and end points of the study period, e.g., c('2019-01-01', '2019-12-31'). Should be specified in valid date format (e.g., 'yyyy-mm-dd').

scu_exclude

(data.frame | data.table) Optional table containing special care unit (SCU) encounters. This is only required if patients who were in an SCU should be excluded from the census calculation during time periods where they were in the SCU. The scu_exclude table typically refers to the ipscu table (see GEMINI database schema). However, users may want to filter the ipscu table by specific SCU unit numbers that should be excluded from the census calculation. Entries where scu_unit_number = 99 ("No SCU") are automatically removed by this function.

The scu_exclude input table needs to contain the following columns:

genc_id (integer): GEMINI encounter ID
scu_admit_date_time (character | POSIXct POSIXt): Date-time of SCU admission in YYYY-MM-DD HH:MM format
scu_discharge_date_time (character | POSIXct POSIXt): Date-time SCU of discharge in YYYY-MM-DD HH:MM format

For all entries in the scu_exclude table that have a valid scu_admit_date_time and scu_discharge_date_time, the encounter will not be counted towards the census during any time periods where they were in the SCU (i.e., time periods where they did not occupy a bed in a non-SCU ward). If no input is provided for this argument, no exclusion will be applied.

group_var

(character) Optional input specifying one (or multiple) grouping variables. Census numbers and capacity ratio will be calculated separately for each level of a given grouping variable. Note: Users do not need to specify hospital as a grouping variable since the function automatically groups the output by hospital_num. Therefore, group_var should only be specified if additional grouping (e.g., by different medical subservices, physicians etc.) is required. Multiple grouping variables can be provided as a character vector (e.g., group_var = c("medical_subservice","physician")). If no grouping variable is specified (default: group_var = NA), the function returns the total daily census numbers and capacity ratios per hospital.

capacity_func

(character) Optional input specifying the function to be used to define capacity for the capacity_ratio output. Can either be a measure of central tendency to obtain typical occupancy (median, mean, or mode) or max to obtain maximum occupancy. The capacity_ratio is calculated as census/capacity_func(census). Default is median (i.e., capacity_ratio = census/median(census)). If "mode" is selected and census has multiple modes, the median of all modes is used to calculate the capacity ratio.

Note that capacity is calculated separately for each site (and grouping variable, if any). Other types of measures (e.g., trimmed means), or year-over-year capacity ratios could be calculated based on the raw census counts in the output (see vignette for more details).

buffer

(integer) Buffer period (in days) to be applied at the end of the data availability period. Default is 30 days, i.e., census counts (and capacity ratio) are set to NA for the last 30 days of available data for each hospital. This is only relevant if the time period of interest is towards the end of the data availability period for a given hospital. For example, if time_period = c("2020-01-01","2021-12-31"), and data are only available until Dec 31 2021 for hospital A, census counts will drop sharply during the last few days in Dec 2021 due to a "truncation bias". Specifically, in the example scenario, there are encounters at hospital A who were admitted prior to Dec 31 2021, but have yet to be discharged. Due to the fact that encounters only appear in the GEMINI database once they have been discharged, the census counts in late December will decline abruptly at hospital A because patients who have not been discharged yet are not included in the database. To prevent this decrease in patient numbers from affecting census counts & capacity ratio, outputs will be set to NA for the last 30 days at hospital A.

The default buffer period is set to 30 because the vast majority of patients is discharged within 30 days. That is, a 30-day buffer period ensures that most admitted patients have already been discharged, and thus, will be included in our data. However, users may specify a different buffer period since length of stays can vary widely between cohorts. Therefore, users should explore which buffer period makes most sense for the data they work with.

Note that data availability differs across hospitals. Therefore, the actual buffer that is applied is calculated separately for each hospital. If a hospital's data availability ends prior to the end of the time period of interest, the buffer includes all last X days of that hospital's data (e.g., 30 days by default). If a hospital's data availability goes past the end of the time period of interest, the actual buffer that is applied is based on the difference between the specified buffer and the latest available discharge date at that hospital. For example, if buffer = 30 and a given hospital has more than 30 extra days of data past the time period of interest, no additional buffer is applied.

time_of_day

(character | numeric) Optional input specifying the time of the day the census/capacity ratio should be calculated at. Default is 8am (time_of_day = "8:00:00"). Other times can be specified as a character (e.g., "10:30" for 10.30am) or as numeric input (e.g., 14 for 2pm).

include_zero

(logical) Flag indicating whether days with census counts of 0 should be treated as real 0s or whether they should be returned as census = NA. Note that this is only relevant for edge cases where this function is applied to small cohorts or if users specify a group_var with small samples resulting in counts of 0 on some days. In those cases, the decision on whether or not to include 0s can affect the estimated typical capacity (and thus, the capacity_ratio output).

By default (include_zero = TRUE), the function will return days where no genc_ids were hospitalized as census = 0. This is the desired behavior in cases where counts of 0 patients are conceptually meaningful. For example, when users want to analyze how many patients with a rare condition are in hospital on a typical day, days where census = 0 represent true counts of 0 patients (assuming the cohort provides sufficient coverage of all hospitalized patients). Note: If data availability periods differ between hospitals, census counts will only be set to 0 for dates that fall within the min-max date range for that particular site.

In other scenarios, users may want to disregard days where census = 0 by specifying include_zero = FALSE. For example, when looking at daily census counts per physician, days where census = 0 likely reflect days where a given physician was not on service. Therefore, those days should not be included in the estimate of physicians' typical patient volume.

Value

(data.table)
data.table with the daily counts of hospitalized patients (census) at each hospital. Additionally, the capacity_ratio (census relative to the typical occupancy during the time period of interest) will be returned. Dates within the time period of interest where no data were available at a given site are not included in the output. For any dates inside the buffer period, census and capacity_ratio are returned as NA.

Examples


## calculate census of all in-patient admissions (ipadm):
if (FALSE) { # \dontrun{
drv <- dbDriver("PostgreSQL")
dbcon <- DBI::dbConnect(drv,
                        dbname = "db",
                        host = "domain_name.ca",
                        port = 1234,
                        user = getPass("Enter user:"),
                        password = getPass("password"))

ipadm <- dbGetQuery(dbcon, "select * from admdad") %>% data.table()

ipadm_census <- daily_census(ipadm)
} # }

Usage

Arguments

Value

See also

Examples