tidier data report, imposed age restriction to 12-80 (configurable)

Snakefile

@@ -34,7 +34,7 @@ rule prepare_data:
     shell:
         """
         mkdir -p output/{wildcards.version}/data
-        Rscript -e "rmarkdown::render(\\"{input.markdown}\\", output_dir = \\"output/{wildcards.version}\\", params = list(datapath = \\"../data/{wildcards.version}\\", max_lab_days = {config[max_lab_days]}, seed = {config[seed]}))"
+        Rscript -e "rmarkdown::render(\\"{input.markdown}\\", output_dir = \\"output/{wildcards.version}\\", params = list(datapath = \\"../data/{wildcards.version}\\", max_lab_days = {config[max_lab_days]}, seed = {config[seed]}, age_min = {config[age_min]}, age_max = {config[age_max]}))"
         mv reports/*.rds output/{wildcards.version}/data
         mv reports/figures.zip {output.figures}
         """
@@ -88,7 +88,7 @@ rule generate_validation_data_v1_1:
           for j in range(1, config["folds"] + 1)
         ]
 
-        
+
 
 
 

config.yml

@@ -2,6 +2,10 @@ seed:
   42
 max_lab_days:
   3
+age_min:
+  12
+age_max:
+  80
 mi_m:
   5
 mi_maxiter:

reports/prepare_data.Rmd

@@ -13,6 +13,8 @@ params:
   datapath: "../data/v1.1"
   max_lab_days: 3
   seed: 42
+  age_min: 12
+  age_max: 80
 ---
 
 
@@ -38,7 +40,9 @@ df_gose     <- readRDS(sprintf('%s/df_gose.rds', params$datapath))
 
 # Extract data
 
-```{r}
+## Baseline and death times
+
+```{r extract-deathtimes}
 df_deaths <- df_baseline %>% 
   transmute(
     gupi, 
@@ -50,7 +54,11 @@ df_deaths <- df_baseline %>%
   ) %>% 
   filter(complete.cases(.))
 ```
-```{r}
+
+We use exact deathtimes (Subject.DeathDate).
+Death dates are recorded for `r df_deaths %>% nrow`.
+
+```{r extract-baseline-covariates}
 df_baseline <- df_baseline %>% 
   select(-Subject.DeathDate) %>% 
   mutate(
@@ -121,13 +129,13 @@ df_baseline <- df_baseline %>%
   )
 ```
 
-
+Overall, `r nrow(df_baseline)` individuals have recorded baseline data.
 
 
 
 ## GOSE data
 
-```{r gose-outcomes-ambiguity}
+```{r extract-gose}
 df_gose <- df_gose %>% 
   distinct %>% 
   filter(complete.cases(.)) %>% 
@@ -136,17 +144,16 @@ df_gose <- df_gose %>%
   mutate(Outcomes.DerivedCompositeGOSE = factor(Outcomes.DerivedCompositeGOSE, levels = 1:8))
 ``` 
 
-This results in `r nrow(df_gose)` GOSE measurements of 
-`r df_gose %>% group_by(gupi) %>% n_groups()` individuals.
-
-
-
-# Compile final datasets
-
-* exclude all patient who do not survive first 6 months (no need to impute)
-* exclude all patients with no GOSE measurement (no imputation)
+Overall, `r nrow(df_gose)` GOSE measurements of 
+`r df_gose %>% group_by(gupi) %>% n_groups()` individuals are available.
+To these observations, we add a GOSE of 1 at the recorded death times.
+We then exclude all patients with a recorded death time of less than 6 months,
+since there 6 months GOSE is known exactly (1, dead).
+The target population is thus the subset of individuals with 
+1. at least one valid GOSE observation
+2. no confirmed death within 6 months
 
-```{r}
+```{r exclude-early-deaths}
 early_deaths_gupi <- df_deaths %>% 
   filter(days <= 180 - 14) %>% 
   .[["gupi"]]
@@ -185,14 +192,14 @@ df_baseline <- df_baseline %>%
 This results in `r nrow(df_gose)` GOSE measurements of 
 `r df_gose %>% group_by(gupi) %>% n_groups()` individuals.
 
-
+ 
 
 ## Plausibility check
 
 The only genuinly numerical variables are Age, Glucose_mmolL, and Hb_dL.
 All other variables are factors and may therefore not contain outliers.
 
-```{r}
+```{r baseline-histograms-raw}
 df_baseline %>% 
   select(Subject.Age, Labs.DLGlucosemmolL, Labs.DLHemoglobingdL) %>% 
   gather(Variable, value) %>% 
@@ -203,19 +210,40 @@ df_baseline %>%
     theme(panel.grid = element_blank())
 ```
 
+
+### Age range
+
+The observed age range is quite wide, we further restrict the study population
+to individuals between `r params$age_min` and `r params$age_max`.
+
+```{r restrict-age-range}
+df_baseline <- df_baseline %>% 
+  filter(
+    Subject.Age >= params$age_min,
+    Subject.Age <= params$age_max
+  )
+df_gose <- df_gose %>% 
+  filter(gupi %in% df_baseline$gupi)
+```
+
+This reduces the number GOSE observations to `r nrow(df_gose)` of
+`r df_gose %>% group_by(gupi) %>% n_groups()` individuals.
+
+
+### Glucose and Hemoglobin
+
 Glucose is obviously left-skewed and a log transfrom might improve fits in linear
 models.
+All values above 50 are considered implausible and set to missing 
+(probably meant as missing).
 
-```{r, echo=TRUE}
+```{r glucose-outliers}
 df_baseline %>% 
   select(gupi, Labs.DLHemoglobingdL) %>% 
   filter(Labs.DLHemoglobingdL > 50)
 ```
 
-All values above 50 are considered implausible and set to missing 
-(probably meant as missing).
-
-```{r}
+```{r log-trans-hemoglobin}
 df_baseline <- df_baseline %>% 
   mutate(
     Labs.DLHemoglobingdL = ifelse(Labs.DLHemoglobingdL > 50, NA_real_, Labs.DLHemoglobingdL),
@@ -334,18 +362,20 @@ ggsave("gose_alluvial_differential_coloring.pdf", height = 5, width = 8)
 ggsave("gose_alluvial_differential_coloring.png", height = 5, width = 8)
 ```
 
+Out of the `r df_gose %>% group_by(gupi) %>% n_groups` individuals in the
+final dataset, `r df_gose %>% group_by(gupi) %>% filter(!any(Outcomes.DerivedCompositeGOSEDaysPostInjury >= 5*30 & Outcomes.DerivedCompositeGOSEDaysPostInjury <= 8*30)) %>% n_groups` do not 
+have per-protocol 6 months GOSE observations and are eligible for model-based
+imputation.
+
 
 
+# Session Info
+
 ```{r zip-figures}
 system("zip figures.zip *.png *.pdf")
 system("rm *.png *.pdf")
 ```
 
-
-
-
-# Session Info
-
 ```{r session-info}
 sessionInfo()
 ```