Continuous Quality Improvement (CQI) and Process Improvement (PI) are essential pillars of healthcare, particularly in the care of injured patients. However, hospitals, trauma systems, and their trauma program managers (TPMs) often lack access to standardized quality measures derived from academic literature. The {traumar} package addresses this gap by providing tools to calculate quality measures related to relative mortality efficiently and accurately. By automating these calculations, {traumar} empowers hospital systems, trauma networks, and TPMs to focus their efforts on analyzing outcomes and driving meaningful improvements in patient care. Whether you’re seeking to enhance PI initiatives or streamline CQI processes, {traumar} serves as a valuable resource for advancing trauma care quality.
You can install the development version of traumar from GitHub with:
# install.packages("pak")
pak::pak("bemts-hhs/traumar"){traumar} has many functions to help you in your data analysis
journey! In particular, if you do not presently have access to
probability of survival data, {traumar} provides the
probability_of_survival() function to do just that using
the TRISS method. Check out the additional package documentation at https://bemts-hhs.github.io/traumar/ where you can find
examples of each function the package has to offer.
The W-Score tells us how many survivals (or deaths) on average out of every 100 cases seen in a trauma center. Using R, we can do this with the {traumar} package.
# Generate example data with high negative skewness
set.seed(123)
# Parameters
n_patients <- 10000 # Total number of patients
# Generate survival probabilities (Ps) using a logistic distribution
Ps <- plogis(rnorm(n_patients, mean = 2, sd = 1.5)) # Skewed towards higher values
# Simulate survival outcomes based on Ps
survival_outcomes <- rbinom(n_patients, size = 1, prob = Ps)
# Create data frame
data <- data.frame(Ps = Ps, survival = survival_outcomes) |>
dplyr::mutate(death = dplyr::if_else(survival == 1, 0, 1))
# Calculate trauma performance (W, M, Z scores)
trauma_performance(data, Ps_col = Ps, outcome_col = death)
#> # A tibble: 9 × 2
#> Calculation_Name Value
#> <chr> <dbl>
#> 1 N_Patients 10000
#> 2 N_Survivors 8137
#> 3 N_Deaths 1863
#> 4 Predicted_Survivors 8097.
#> 5 Predicted_Deaths 1903.
#> 6 Patient_Estimate 40.3
#> 7 W_Score 0.403
#> 8 M_Score 0.374
#> 9 Z_Score 1.18The M and Z scores are calculated using methods defined in the literature
may not be meaningful if your the distribution of the probability of
survival measure is not similar enough to the Major Trauma Outcomes
Study distribution. {traumar} provides a way to check this in your data
analysis script, or even from the console. The
trauma_performance() function does this under the hood for
you, so you can get a read out of how much confidence you can put into
the Z score.
# Compare the current case mix with the MTOS case mix
trauma_case_mix(data, Ps_col = Ps, outcome_col = death)
#> Ps_range current_fraction MTOS_distribution
#> 1 0.00 - 0.25 0.0209 0.010
#> 2 0.26 - 0.50 0.0742 0.043
#> 3 0.51 - 0.75 0.1907 0.000
#> 4 0.76 - 0.90 0.3000 0.052
#> 5 0.91 - 0.95 0.1979 0.053
#> 6 0.96 - 1.00 0.2163 0.842Napoli et al.(2017) published methods for calculating a measure of trauma center (or system) performance while overcoming a problem with the W-Score and the TRISS methodology. Given that the majority of patients seen at trauma centers will have a probability of survival over 90%, estimating performance based on the W-Score may only indicate how well a center performed with lower acuity patients. Using Napoli et al. (2017), it is possible to calculate a score that is similar to the W-Score in its interpretability, but deals with the negatively skewed probability of survival problem by creating non-linear bins of score ranges, and then weighting a score based on the nature of those bins. The Relative Mortality Metric (RMM) has a scale from -1 to 1, where
An important part of the approach Napoli et al. (2017) took was to modify the M-Score approach of looking at linear bins of the probability of survival distribution, and make it non-linear. The {traumar} package does this for you using Dr. Napoli’s method:
# Apply the nonlinear_bins function
results <- nonlinear_bins(data = data,
Ps_col = Ps,
outcome_col = survival,
divisor1 = 4,
divisor2 = 4,
threshold_1 = 0.9,
threshold_2 = 0.99)
# View intervals created by the algorithm
results$intervals
#> [1] 0.02257717 0.59018811 0.75332640 0.84397730 0.90005763 0.93040607
#> [7] 0.95446838 0.97345230 0.99000626 0.99957866
# View the bin statistics
results$bin_stats
#> # A tibble: 9 × 13
#> bin_number bin_start bin_end mean sd Pred_Survivors_b Pred_Deaths_b
#> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 1 0.0226 0.590 0.416 0.133 577. 812.
#> 2 2 0.590 0.753 0.681 0.0480 946. 443.
#> 3 3 0.753 0.844 0.803 0.0261 1116. 273.
#> 4 4 0.844 0.900 0.873 0.0162 1211. 176.
#> 5 5 0.900 0.930 0.916 0.00879 921. 84.5
#> 6 6 0.930 0.954 0.943 0.00699 949. 57.3
#> 7 7 0.954 0.973 0.964 0.00545 970. 36.0
#> 8 8 0.973 0.990 0.981 0.00485 987. 18.7
#> 9 9 0.990 1.00 0.994 0.00253 419. 2.61
#> # ℹ 6 more variables: AntiS_b <dbl>, AntiM_b <dbl>, alive <dbl>, dead <dbl>,
#> # count <dbl>, percent <dbl>The RMM is sensitive to higher acuity patients, meaning that if a
trauma center struggles with these patients, it will be reflected in the
RMM. In contrast, the W-Score may mask declines in performance due to
the influence of lower acuity patients via the MTOS Distribution. The
{traumar} package automates RMM calculation as a single score using the
nonlinear binning method from Napoli et al. (2017). The
rmm() and rm_bin_summary() functions
internally call nonlinear_bins() to generate the non-linear
binning process. The function uses a bootstrap process with
n_samples repetitions to simulate an RMM distribution and
estimate 95% confidence intervals. The RMM, along with corresponding
confidence intervals, are provided for the population in
data, as well.
# Example usage of the `rmm()` function
rmm(data = data,
Ps_col = Ps,
outcome_col = survival,
n_samples = 250,
Divisor1 = 4,
Divisor2 = 4
)
#> # A tibble: 1 × 8
#> population_RMM_LL population_RMM population_RMM_UL population_CI
#> <dbl> <dbl> <dbl> <dbl>
#> 1 -0.0280 0.0365 0.101 0.0645
#> # ℹ 4 more variables: bootstrap_RMM_LL <dbl>, bootstrap_RMM <dbl>,
#> # bootstrap_RMM_UL <dbl>, bootstrap_CI <dbl>
# Pivoting can be helpful at times
rmm(
data = data,
Ps_col = Ps,
outcome_col = survival,
n_samples = 250,
Divisor1 = 4,
Divisor2 = 4,
pivot = TRUE
)
#> # A tibble: 8 × 2
#> stat value
#> <chr> <dbl>
#> 1 population_RMM_LL -0.0280
#> 2 population_RMM 0.0365
#> 3 population_RMM_UL 0.101
#> 4 population_CI 0.0645
#> 5 bootstrap_RMM_LL 0.0329
#> 6 bootstrap_RMM 0.0353
#> 7 bootstrap_RMM_UL 0.0378
#> 8 bootstrap_CI 0.00244
# RMM calculated by non-linear bin range
# `rm_bin_summary()` function
rm_bin_summary(data = data,
Ps_col = Ps,
outcome_col = survival,
Divisor1 = 4,
Divisor2 = 4,
n_samples = 250
)
#> # A tibble: 9 × 19
#> bin_number TA_b TD_b N_b EM_b AntiS_b AntiM_b bin_start bin_end
#> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 1 614 775 1389 0.558 0.416 0.584 0.0226 0.590
#> 2 2 953 436 1389 0.314 0.681 0.319 0.590 0.753
#> 3 3 1108 281 1389 0.202 0.803 0.197 0.753 0.844
#> 4 4 1208 179 1387 0.129 0.873 0.127 0.844 0.900
#> 5 5 911 95 1006 0.0944 0.916 0.084 0.900 0.930
#> 6 6 954 52 1006 0.0517 0.943 0.057 0.930 0.954
#> 7 7 979 27 1006 0.0268 0.964 0.036 0.954 0.973
#> 8 8 989 17 1006 0.0169 0.981 0.019 0.973 0.990
#> 9 9 421 1 422 0.00237 0.994 0.006 0.990 1.00
#> # ℹ 10 more variables: midpoint <dbl>, R_b <dbl>, population_RMM_LL <dbl>,
#> # population_RMM <dbl>, population_RMM_UL <dbl>, population_CI <dbl>,
#> # bootstrap_RMM_LL <dbl>, bootstrap_RMM <dbl>, bootstrap_RMM_UL <dbl>,
#> # bootstrap_CI <dbl>