update project design

2026-06-03 22:06:23 -08:00
parent 49409a7d1b
commit 73e5d46c30
27 changed files with 490 additions and 8513 deletions
@@ -0,0 +1,164 @@
+---
+title: "Explore Simulated ICU Data"
+format:
+  html:
+    embed-resources: true
+execute:
+  echo: true
+  warning: false
+  message: false
+---
+
+## Goal
+
+This notebook is for exploring the simulated ICU cohort.
+
+The report notebook focuses on target trial components and core results. This exploratory notebook focuses on visualizing the data-generating process and understanding why the naive comparison can be biased.
+
+## Setup
+
+```{r}
+suppressPackageStartupMessages({
+  library(dplyr)
+  library(ggplot2)
+  library(gt)
+  library(gtsummary)
+  library(tibble)
+})
+
+source("../R/simulate_icu_cohort.R")
+source("../R/estimate_naive_vasopressor_mortality_effect.R")
+```
+
+## Simulate Data
+
+```{r}
+icu_data <- simulate_icu_cohort(n_patients = 1000, seed = 20260531)
+naive_mortality_analysis <- estimate_naive_vasopressor_mortality_effect(icu_data)
+eligible_icu_patients <- naive_mortality_analysis$eligible_icu_patients
+```
+
+## Eligibility Overview
+
+```{r}
+eligibility_overview <- icu_data |>
+  summarize(
+    simulated_patients = n(),
+    suspected_sepsis = sum(suspected_sepsis == 1),
+    hypotension_at_baseline = sum(hypotension_at_baseline == 1),
+    elevated_lactate_at_baseline = sum(elevated_lactate_at_baseline == 1),
+    eligible = sum(eligible == 1)
+  ) |>
+  tidyr::pivot_longer(
+    cols = everything(),
+    names_to = "measure",
+    values_to = "patients"
+  )
+
+eligibility_overview |>
+  gt() |>
+  tab_header(title = "Eligibility Overview") |>
+  cols_label(
+    measure = "Measure",
+    patients = "Patients"
+  ) |>
+  fmt_integer(columns = patients)
+```
+
+## Baseline Distributions
+
+```{r}
+eligible_icu_patients |>
+  select(age, sofa_score, lactate, map) |>
+  tbl_summary(
+    statistic = all_continuous() ~ "{mean} ({sd}); {median} [{p25}, {p75}]",
+    missing = "no"
+  )
+```
+
+## Severity By Observed Treatment
+
+```{r}
+eligible_icu_patients |>
+  mutate(
+    observed_treatment = factor(
+      early_vasopressor,
+      levels = c(0, 1),
+      labels = c("No early vasopressor", "Early vasopressor")
+    )
+  ) |>
+  ggplot(aes(x = observed_treatment, y = sofa_score, fill = observed_treatment)) +
+  geom_boxplot(alpha = 0.75, width = 0.65, show.legend = FALSE) +
+  labs(
+    title = "SOFA Score Is Higher In Early-Treated Patients",
+    x = NULL,
+    y = "SOFA score"
+  ) +
+  theme_minimal()
+```
+
+Early vasopressor patients tend to have higher SOFA scores in this simulated cohort. That happens because the treatment assignment mechanism makes sicker patients more likely to receive early vasopressors.
+
+## Lactate And MAP By Treatment
+
+```{r}
+eligible_icu_patients |>
+  mutate(
+    observed_treatment = factor(
+      early_vasopressor,
+      levels = c(0, 1),
+      labels = c("No early vasopressor", "Early vasopressor")
+    )
+  ) |>
+  ggplot(aes(x = map, y = lactate, color = observed_treatment)) +
+  geom_point(alpha = 0.55) +
+  labs(
+    title = "Shock Severity Markers Differ By Observed Treatment",
+    x = "Mean arterial pressure",
+    y = "Lactate",
+    color = "Observed treatment"
+  ) +
+  theme_minimal()
+```
+
+Higher lactate and lower MAP are both markers of greater shock severity. If treatment groups differ on these variables, a simple treated-versus-untreated comparison is not yet a causal estimate.
+
+## Mortality By SOFA Score
+
+```{r}
+eligible_icu_patients |>
+  mutate(
+    observed_treatment = factor(
+      early_vasopressor,
+      levels = c(0, 1),
+      labels = c("No early vasopressor", "Early vasopressor")
+    )
+  ) |>
+  ggplot(aes(x = sofa_score, y = death_28d, color = observed_treatment)) +
+  geom_jitter(height = 0.04, width = 0.15, alpha = 0.35) +
+  geom_smooth(method = "glm", method.args = list(family = "binomial"), se = FALSE) +
+  labs(
+    title = "Mortality Risk Rises With SOFA Score",
+    x = "SOFA score",
+    y = "Observed 28-day death",
+    color = "Observed treatment"
+  ) +
+  theme_minimal()
+```
+
+This plot shows why baseline severity matters. If SOFA score predicts death and also affects treatment assignment, then SOFA score is a confounder for the naive treatment comparison.
+
+## Naive Mortality Effect Estimates
+
+```{r}
+naive_mortality_analysis$mortality_effect_estimates |>
+  gt() |>
+  tab_header(title = "Naive 28-Day Mortality Effect Estimates") |>
+  cols_label(
+    estimate = "Estimate",
+    value = "Value"
+  ) |>
+  fmt_number(columns = value, decimals = 3)
+```
+
+The next methodological step is to adjust for baseline severity rather than comparing observed treatment groups directly.