Predictive question-1

# Load required packages
require(tableone)
require(Publish)
require(MatchIt)
require(cobalt)
require(ggplot2)

Working with a Predictive question using RHC

This tutorial delves into processing and understanding the right heart catheterization (RHC) dataset, which pertains to patients in the intensive care unit. The dataset is particularly centered around the implications of using RHC in the early phases of care, with a focus on comparing two patient groups: those who received the RHC procedure and those who did not. The key outcome being analyzed is the 30-day survival rate. We will use this as an example to explain how to work with a predictive research question to build the analytic data.

Link for the RHC dataset

(Connors et al. 1996) published an article in JAMA. The article is about managing or guiding therapy for the critically ill patients in the intensive care unit. They considered a number of health-outcomes such as

  • length of stay (hospital stay; measured continuously)
  • death within certain period (death at any time up to 180 Days; measured as a binary variable)

The original article was concerned about the association of right heart catheterization (RHC) use during the first 24 hours of care in the intensive care unit and the health-outcomes mentioned above.

But we will use this data as a case study for our prediction modelling. Traditional PICOT framework is designed primarily for clinical questions related to interventions, so when applying it to other areas like predictive modeling, some creative adaptation is needed.

Aspect Description
P Patients who are critically ill
I Not applicable, as we are dealing with a prediction model here
C Not applicable, as we are dealing with a prediction model here
O in-hospital mortality
T Between 1989 to 1994 (see the JAMA paper)

We are interested in developing a prediction model for the length of stay.

Data download

Data is freely available from Vanderbilt Biostatistics, variable list is available here, and the article is freely available from researchgate.

RHC Data amd search for right heart catheterization dataset

Variable list

Article

Let us download the dataset and save it for later use.

# Load the dataset
ObsData <- read.csv("https://hbiostat.org/data/repo/rhc.csv", 
                    header = TRUE)

# Save the dataset
saveRDS(ObsData, file = "Data/researchquestion/rhc.RDS")

Creating analytic dataset

Now, we show the process of preparing our analytic dataset (i.e., ready to use dataset for our analysis), so that the variables generally match with the way the authors were coded in the original article. Below we show the process of creating the analytic dataset.

Add column for outcome: length of stay

# Length of Stay = date of discharge - study admission date
ObsData$Length.of.Stay <- ObsData$dschdte - ObsData$sadmdte

# Length of Stay = date of death - study admission date if date of discharge not available
ObsData$Length.of.Stay[is.na(ObsData$Length.of.Stay)] <- 
  ObsData$dthdte[is.na(ObsData$Length.of.Stay)] - 
  ObsData$sadmdte[is.na(ObsData$Length.of.Stay)]

Recoding column for outcome: death

Tip

Here we use the ifelse function to create a categorical variable. Other related functions are cut, car.

Let us recode our outcome variable as a binary variable:

ObsData$death <- ifelse(ObsData$death == "Yes", 1, 0)

Remove unnecessary outcomes

Our next task is to remove unnecessary outcomes:

Tip

There are multiple ways to drop variables from a dataset. E.g., without using any package and using the select function from the dplyr package.

ObsData <- dplyr::select(ObsData, !c(dthdte, lstctdte, dschdte, 
                            t3d30, dth30, surv2md1))

Remove unnecessary and problematic variables

Now we will drop unnecessary and problematic variables:

ObsData <- dplyr::select(ObsData, !c(sadmdte, ptid, X, adld3p, 
                                     urin1, cat2))

Basic data cleanup

Now we will do some basic cleanup.

Tip

We an use the lapply function to convert all categorical variables to factors at once. Not that a similar function to lapply is sapply. The main difference is that sapply attempts to convert the result into a vector or matrix, while lapply returns a list.

# convert all categorical variables to factors
factors <- c("cat1", "ca", "death", "cardiohx", "chfhx", 
             "dementhx", "psychhx", "chrpulhx", "renalhx", 
             "liverhx", "gibledhx", "malighx", "immunhx", 
             "transhx", "amihx", "sex", "dnr1", "ninsclas", 
             "resp", "card", "neuro", "gastr", "renal", "meta", 
             "hema", "seps", "trauma", "ortho", "race", 
             "income")
ObsData[factors] <- lapply(ObsData[factors], as.factor)

# convert RHC.use (RHC vs. No RHC) to a binary variable
ObsData$RHC.use <- ifelse(ObsData$swang1 == "RHC", 1, 0)
ObsData <- dplyr::select(ObsData, !swang1)

# Categorize the variables to match with the original paper
ObsData$age <- cut(ObsData$age, breaks=c(-Inf, 50, 60, 70, 80, Inf),
                   right=FALSE)
ObsData$race <- factor(ObsData$race, 
                       levels=c("white","black","other"))
ObsData$sex <- as.factor(ObsData$sex)
ObsData$sex <- relevel(ObsData$sex, ref = "Male")
ObsData$cat1 <- as.factor(ObsData$cat1)
levels(ObsData$cat1) <- c("ARF","CHF","Other","Other","Other",
                          "Other","Other","MOSF","MOSF")
ObsData$ca <- as.factor(ObsData$ca)
levels(ObsData$ca) <- c("Metastatic","None","Localized (Yes)")
ObsData$ca <- factor(ObsData$ca, levels=c("None", "Localized (Yes)",
                                          "Metastatic"))

Rename variables

# Rename the variables
names(ObsData) <- c("Disease.category", "Cancer", "Death", 
                    "Cardiovascular", "Congestive.HF", 
                    "Dementia", "Psychiatric", "Pulmonary", 
                    "Renal", "Hepatic", "GI.Bleed", "Tumor", 
                    "Immunosupperssion", "Transfer.hx", "MI", 
                    "age", "sex", "edu", "DASIndex", 
                    "APACHE.score", "Glasgow.Coma.Score", 
                    "blood.pressure", "WBC", "Heart.rate",
                    "Respiratory.rate",  "Temperature",
                    "PaO2vs.FIO2", "Albumin", "Hematocrit", 
                    "Bilirubin", "Creatinine", "Sodium", 
                    "Potassium", "PaCo2",  "PH", "Weight", 
                    "DNR.status", "Medical.insurance", 
                    "Respiratory.Diag", "Cardiovascular.Diag", 
                    "Neurological.Diag", "Gastrointestinal.Diag",
                    "Renal.Diag", "Metabolic.Diag", 
                    "Hematologic.Diag", "Sepsis.Diag", 
                    "Trauma.Diag", "Orthopedic.Diag", 
                    "race", "income", 
                    "Length.of.Stay", "RHC.use")

# Save the dataset
saveRDS(ObsData, file = "Data/researchquestion/rhcAnalytic.RDS")

Notations

let us introduce with some notations:

Notations Example in RHC study
\(Y_1\): Observed outcome length of stay
\(Y_2\): Observed outcome death within 3 months
\(L\): Covariates See below

Basic data exploration

Dimension

Let us the how many rows and columns we have:

dim(ObsData)
#> [1] 5735   52

Comprehensive summary

Let us see the summary statistics of the variables:

Tip

To see the comprehensive summary of the variables, we can use the skim function form skimr package or describe function from rms package

require(skimr)
skim(ObsData)
Data summary
Name ObsData
Number of rows 5735
Number of columns 52
_______________________
Column type frequency:
factor 31
numeric 21
________________________
Group variables None

Variable type: factor

skim_variable n_missing complete_rate ordered n_unique top_counts
Disease.category 0 1 FALSE 4 ARF: 2490, MOS: 1626, Oth: 1163, CHF: 456
Cancer 0 1 FALSE 3 Non: 4379, Loc: 972, Met: 384
Death 0 1 FALSE 2 1: 3722, 0: 2013
Cardiovascular 0 1 FALSE 2 0: 4722, 1: 1013
Congestive.HF 0 1 FALSE 2 0: 4714, 1: 1021
Dementia 0 1 FALSE 2 0: 5171, 1: 564
Psychiatric 0 1 FALSE 2 0: 5349, 1: 386
Pulmonary 0 1 FALSE 2 0: 4646, 1: 1089
Renal 0 1 FALSE 2 0: 5480, 1: 255
Hepatic 0 1 FALSE 2 0: 5334, 1: 401
GI.Bleed 0 1 FALSE 2 0: 5550, 1: 185
Tumor 0 1 FALSE 2 0: 4419, 1: 1316
Immunosupperssion 0 1 FALSE 2 0: 4192, 1: 1543
Transfer.hx 0 1 FALSE 2 0: 5073, 1: 662
MI 0 1 FALSE 2 0: 5535, 1: 200
age 0 1 FALSE 5 [-I: 1424, [60: 1389, [70: 1338, [50: 917
sex 0 1 FALSE 2 Mal: 3192, Fem: 2543
DNR.status 0 1 FALSE 2 No: 5081, Yes: 654
Medical.insurance 0 1 FALSE 6 Pri: 1698, Med: 1458, Pri: 1236, Med: 647
Respiratory.Diag 0 1 FALSE 2 No: 3622, Yes: 2113
Cardiovascular.Diag 0 1 FALSE 2 No: 3804, Yes: 1931
Neurological.Diag 0 1 FALSE 2 No: 5042, Yes: 693
Gastrointestinal.Diag 0 1 FALSE 2 No: 4793, Yes: 942
Renal.Diag 0 1 FALSE 2 No: 5440, Yes: 295
Metabolic.Diag 0 1 FALSE 2 No: 5470, Yes: 265
Hematologic.Diag 0 1 FALSE 2 No: 5381, Yes: 354
Sepsis.Diag 0 1 FALSE 2 No: 4704, Yes: 1031
Trauma.Diag 0 1 FALSE 2 No: 5683, Yes: 52
Orthopedic.Diag 0 1 FALSE 2 No: 5728, Yes: 7
race 0 1 FALSE 3 whi: 4460, bla: 920, oth: 355
income 0 1 FALSE 4 Und: 3226, $11: 1165, $25: 893, > $: 451

Variable type: numeric

skim_variable n_missing complete_rate mean sd p0 p25 p50 p75 p100 hist
edu 0 1 11.68 3.15 0.00 10.00 12.00 13.00 30.00 ▁▇▃▁▁
DASIndex 0 1 20.50 5.32 11.00 16.06 19.75 23.43 33.00 ▃▇▆▂▃
APACHE.score 0 1 54.67 19.96 3.00 41.00 54.00 67.00 147.00 ▂▇▅▁▁
Glasgow.Coma.Score 0 1 21.00 30.27 0.00 0.00 0.00 41.00 100.00 ▇▂▂▁▁
blood.pressure 0 1 78.52 38.05 0.00 50.00 63.00 115.00 259.00 ▆▇▆▁▁
WBC 0 1 15.65 11.87 0.00 8.40 14.10 20.05 192.00 ▇▁▁▁▁
Heart.rate 0 1 115.18 41.24 0.00 97.00 124.00 141.00 250.00 ▁▂▇▂▁
Respiratory.rate 0 1 28.09 14.08 0.00 14.00 30.00 38.00 100.00 ▅▇▂▁▁
Temperature 0 1 37.62 1.77 27.00 36.09 38.09 39.00 43.00 ▁▁▅▇▁
PaO2vs.FIO2 0 1 222.27 114.95 11.60 133.31 202.50 316.62 937.50 ▇▇▁▁▁
Albumin 0 1 3.09 0.78 0.30 2.60 3.50 3.50 29.00 ▇▁▁▁▁
Hematocrit 0 1 31.87 8.36 2.00 26.10 30.00 36.30 66.19 ▁▆▇▃▁
Bilirubin 0 1 2.27 4.80 0.10 0.80 1.01 1.40 58.20 ▇▁▁▁▁
Creatinine 0 1 2.13 2.05 0.10 1.00 1.50 2.40 25.10 ▇▁▁▁▁
Sodium 0 1 136.77 7.66 101.00 132.00 136.00 142.00 178.00 ▁▂▇▁▁
Potassium 0 1 4.07 1.03 1.10 3.40 3.80 4.60 11.90 ▂▇▁▁▁
PaCo2 0 1 38.75 13.18 1.00 31.00 37.00 42.00 156.00 ▃▇▁▁▁
PH 0 1 7.39 0.11 6.58 7.34 7.40 7.46 7.77 ▁▁▂▇▁
Weight 0 1 67.83 29.06 0.00 56.30 70.00 83.70 244.00 ▂▇▁▁▁
Length.of.Stay 0 1 21.56 25.87 2.00 7.00 14.00 25.00 394.00 ▇▁▁▁▁
RHC.use 0 1 0.38 0.49 0.00 0.00 0.00 1.00 1.00 ▇▁▁▁▅

Predictive vs. causal models

The focus of current document is predictive models (e.g., predicting a health outcome).

The original article by Connors et al. (1996) focused on the association of

Connors et al. (1996)

  • right heart catheterization (RHC) use during the first 24 hours of care in the intensive care unit (exposure of primary interest) and
  • the health-outcomes (such as length of stay).

Then the PICOT table changes as follows:

Aspect Description
P Patients who are critically ill
I Receiving a right heart catheterization (RHC)
C Not receiving a right heart catheterization (RHC)
O length of stay
T Between 1989 to 1994 (see the JAMA paper)

Video content (optional)

Tip

For those who prefer a video walkthrough, feel free to watch the video below, which offers a description of an earlier version of the above content.

References

Connors, Alfred F, Theodore Speroff, Neal V Dawson, Charles Thomas, Frank E Harrell, Douglas Wagner, Norman Desbiens, et al. 1996. “The Effectiveness of Right Heart Catheterization in the Initial Care of Critically III Patients.” Jama 276 (11): 889–97. https://tinyurl.com/Connors1996.