4 Results of IDA: Missing values

This section reports the IDA analyses for missing data. The Section headers (i.e. M2) correspond to the IDA analysis plan in Chapter 2.

4.1 M1: Participant missingness

As the data was obtained from the registry of the laboratory, and only performed laboratory analyses were included, participant missingness cannot be evaluated.

4.2 M2: Variable missingness

Number and percentage of missingness for each predictor, sorted by descending missingness proportion.

4.2.1 Outcome and Structural variables

Variable	Missing (count)	Missing (%)
Summary of missing values for outcome and structural predictors
BACTEREMIA	0	0.0
AGE	0	0.0
SEX	0	0.0

4.2.2 Lab parameters

Missingness patterns for key and medium importance predictors are reported here. The remaining predictors are reported in Appendix Section D.1 .

Predictor	Description	Missing (count)	Missing (%)
Summary of missing values for key predictors
NEU	Neutrophiles (G/L)	728	5.0
WBC	White blood count (G/L)	462	3.1
BUN	Blood urea nitrogen (mg/dl)	172	1.2
CREA	Creatinine (mg/dl)	159	1.1
PLT	Blood platelets (G/L)	42	0.3
AGE	Patient Age (years)	0	0.0

Predictor	Description	Missing (count)	Missing (%)
Summary of missing values for medium importance predictors
FIB	Fibrinogen (mg/dl)	2567	17.5
POTASS	Potassium (mmol/L)	2008	13.7
GGT	Gamma-glutamyl transpeptidase (G/L)	1262	8.6
ASAT	Aspartate transaminase (U/L)	1154	7.9
ALAT	Alanin transaminase (U/L)	987	6.7
CRP	C-reactive protein (mg/dl)	155	1.1

4.3 M3: Complete cases

Number of available complete cases (outcome and predictors):

Set	Complete (count)	Complete (%)
Summary of complete cases by sets of predictors
Outcome	14691	100.0
Outcome and structural variables	14691	100.0
Outcome and key predictors only	13793	93.9
Outcome key predictors and predictors of medium importance	9389	63.9
Outcome and all predictors	3979	27.1

4.4 M4: Patterns of missing values

4.4.1 Complete cases by strata defined by structural variables

Set	# patients	Complete (count)	Complete (%)
Summary of complete cases by sets of predictors
female - [16, 50]
All predictors	2462	604	24.5
Key predictors	2462	2309	93.8
Medium importance predictors	2462	1656	67.3
female - (50, 65]
All predictors	1579	411	26.0
Key predictors	1579	1468	93.0
Medium importance predictors	1579	1075	68.1
female - (65, 101]
All predictors	2114	545	25.8
Key predictors	2114	1975	93.4
Medium importance predictors	2114	1389	65.7
male - [16, 50]
All predictors	2903	805	27.7
Key predictors	2903	2744	94.5
Medium importance predictors	2903	1993	68.7
male - (50, 65]
All predictors	2671	771	28.9
Key predictors	2671	2504	93.7
Medium importance predictors	2671	1862	69.7
male - (65, 101]
All predictors	2962	843	28.5
Key predictors	2962	2793	94.3
Medium importance predictors	2962	2014	68.0

4.4.2 Dendrogram of missingness indicators

The dendrogram depicts the results of a cluster analysis using the complete linkage method based on the percentage of discordant missing indicators. (This percentage was computed via the squared Euclidian distance of missingness indicators between predictors.) The horizontal axis shows the distance between two clusters, which is given by the maximum distance between any element of the first and the second clusters. For example, if two clusters are merged at a height of 25 it means that in 25% of the observations the missingness indicators of the most discordant predictors contained in the two clusters are discordant.

The numbers in brackets are the percentages of missing observations for each predictor.

# Results of IDA: Missing values {#Missing} ```{r, warning=FALSE, message=FALSE} ## Load libraries library(here) library(dplyr) library(tidyr) library(naniar) ## for missing data library(gt) library(gtExtras) library(dendextend) # Load the first iteration of the IDA ready data sets ADSL <- readRDS(here::here("data", "IDA", "ADSL_01.rds")) ADLB <- readRDS(here::here("data", "IDA", "ADLB_01.rds")) ``` This section reports the IDA analyses for missing data. The Section headers (i.e. M2) correspond to the IDA analysis plan in @sec-IDA_plan. ## M1: Participant missingness As the data was obtained from the registry of the laboratory, and only performed laboratory analyses were included, participant missingness cannot be evaluated. ## M2: Variable missingness Number and percentage of missingness for each predictor, sorted by descending missingness proportion. ### Outcome and Structural variables ```{r m1-structural, message = FALSE, warning = FALSE, echo = FALSE} ADSL |> select(BACTEREMIA, AGE, SEX) |> miss_var_summary() |> gt() |> gt::cols_label( variable = "Variable", n_miss = "Missing (count)", pct_miss = "Missing (%)" ) |> fmt_number( columns = vars(pct_miss), decimals = 1 ) |> tab_header(title = "Summary of missing values for outcome and structural predictors") |> gt_theme_538() ``` ### Lab parameters Missingness patterns for key and medium importance predictors are reported here. The remaining predictors are reported in Appendix @sec-M1-appendix . ```{r m1-lab-predictors-key, message = FALSE, warning = FALSE, echo = FALSE} source(here("R", "M1-predictor-missing.R")) table_pred_missing_by_type(data = ABLB, flag = "KEY_PRED_FL01", tab_title = "Summary of missing values for key predictors") ``` ```{r m1-lab-predictors-med, message = FALSE, warning = FALSE, echo = FALSE} table_pred_missing_by_type(data = ABLB, flag = "MED_PRED_FL01", tab_title = "Summary of missing values for medium importance predictors") ``` ## M3: Complete cases Number of available complete cases (outcome and predictors): ```{r summiss, message =FALSE, warning =FALSE , echo=FALSE} source(here("R", "M2-complete-cases.R")) m2_ard <- m2_analysis(ADSL = ADSL, ADLB = ADLB) m2_ard |> select(Set, n_cases, pct_cases ) |> gt() |> cols_label( Set = "Set", n_cases = "Complete (count)", pct_cases = "Complete (%)" ) |> fmt_number( columns = c(pct_cases), decimals = 1 ) |> tab_header(title = "Summary of complete cases by sets of predictors") |> gt_theme_538() ``` ## M4: Patterns of missing values ### Complete cases by strata defined by structural variables ```{r summiss2, message =FALSE, warning =FALSE , echo=FALSE} source(here("R", "M3-complete-cases-sets.R")) m3_ard <- m3_analysis(ADSL, ADLB) m3_ard |> arrange(SEXC, AGEGR01) |> select(Set, n, n_cases, pct_cases, SEXC, AGEGR01C ) |> group_by(SEXC, AGEGR01C) |> gt() |> gt::cols_label( Set = "Set", n = "# patients", n_cases = "Complete (count)", pct_cases = "Complete (%)" ) |> gt::fmt_number( columns = c(pct_cases), decimals = 1 ) |> tab_header(title = "Summary of complete cases by sets of predictors") |> gt_theme_538() ``` ### Dendrogram of missingness indicators The dendrogram depicts the results of a cluster analysis using the complete linkage method based on the percentage of discordant missing indicators. (This percentage was computed via the squared Euclidian distance of missingness indicators between predictors.) The horizontal axis shows the distance between two clusters, which is given by the maximum distance between any element of the first and the second clusters. For example, if two clusters are merged at a height of 25 it means that in 25% of the observations the missingness indicators of the most discordant predictors contained in the two clusters are discordant. The numbers in brackets are the percentages of missing observations for each predictor. ```{r m3-dendogram, message =FALSE, warning =FALSE , echo=FALSE, fig.height=13, fig.width=13} #fig.cap="Clustered variables by percentage observations concordantly missing [by variable percentage missing]" source(here("R", "M3-dendogram.R")) m3_dendogram(ADLB) ```