Compute standardized mean differences ref #78

[tompollard]

This change adds an smd argument to compute pairwise standardized mean differences for continuous and categorical variables. Comparisons with the R package are shown below.

PhysioNet 2012 demo data

R

library(tidyverse)
library(tableone)

path = "https://raw.githubusercontent.com/tompollard/tableone/master/data/pn2012_demo.csv"

pn = read.csv(path)

cols = c("Age","SysABP","Height","Weight","ICU","MechVent","LOS","death")
strata = "MechVent"
tabUnmatched <- CreateTableOne(vars = cols, strata = strata, data = pn, test = FALSE)

print(tabUnmatched, smd = TRUE)

Python

path = "https://raw.githubusercontent.com/tompollard/tableone/master/data/pn2012_demo.csv"

df = pd.read_csv(path)

cols = ["Age", "SysABP", "Height", "Weight", "ICU", "MechVent", "LOS", "death"]
categorical=["ICU", "MechVent", "death"]
strata = "MechVent"

t = TableOne(df, categorical=categorical, label_suffix=True, groupby=strata, 
    pval=True, pval_test_name=False, smd=True)

Right Heart Catheterization Dataset

R

library(tableone)

path = "http://biostat.mc.vanderbilt.edu/wiki/pub/Main/DataSets/rhc.csv"
df = read.csv(path)

vars <- c("age","sex","race","edu","income","ninsclas","cat1","das2d3pc",
          "dnr1","ca","surv2md1","aps1","scoma1","wtkilo1","temp1","meanbp1",
          "resp1","hrt1","pafi1","paco21","ph1","wblc1","hema1","sod1","pot1",
          "crea1","bili1","alb1","resp","card","neuro","gastr","renal",
          "meta","hema","seps","trauma","ortho","cardiohx","chfhx",
          "dementhx","psychhx","chrpulhx","renalhx","liverhx","gibledhx",
          "malighx","immunhx", "transhx","amihx")

tabUnmatched <- CreateTableOne(vars = vars, strata = "swang1", data = rhc, test = FALSE)

print(tabUnmatched, smd = TRUE)

Python

path = "http://biostat.mc.vanderbilt.edu/wiki/pub/Main/DataSets/rhc.csv"
df = pd.read_csv(path)

cols = ["age","sex","race","edu","income","ninsclas","cat1","das2d3pc","dnr1",
          "ca","surv2md1","aps1","scoma1","wtkilo1","temp1","meanbp1","resp1",
          "hrt1","pafi1","paco21","ph1","wblc1","hema1","sod1","pot1","crea1",
          "bili1","alb1","resp","card","neuro","gastr","renal","meta","hema",
          "seps","trauma","ortho","cardiohx","chfhx","dementhx","psychhx",
          "chrpulhx","renalhx","liverhx","gibledhx","malighx","immunhx",
          "transhx","amihx"]

cats = ["race", "income", "ninsclas", "cat1", "dnr1", "ca", "resp",
        "card", "neuro", "gastr", "renal", "meta", "hema", "seps", "trauma",
        "ortho", "sex"]

t = TableOne(df, columns=cols, categorical=cats, label_suffix=True, groupby="swang1", 
             pval=False, pval_test_name=False, smd=True)
t

[jraffa]

Looks good. I'll take a closer look later. Any thoughts on the +/- compared to R?

[tompollard]

Any thoughts on the +/- compared to R?

Good point, I wasn't sure how best to handle this so it would be interesting to hear your thoughts. It feels like the sign offers some useful information, so I prefer not to remove it unless it is confusing.

I've given the name of the two groups in the column header (unlike R) which makes the sign more meaningful, but it may not be clear to the user that the left group is the control and the right is the treatment (i.e. SMD (0,1) means that 0 is the control and 1 is the treatment).

[tompollard]

While I think of it, I wasn't clear whether Hedges correction was valid for the categorical SMD. It doesn't really matter because the option for applying the correction isn't offered, but it would be good to know.

[raheems]

Thanks for the updates. Is it available yet in PyPI?

[tompollard]

@raheems not yet, but it should be shortly (after a little more testing, and hopefully feedback from @jraffa)

[raheems]

I was able to use the smd feature. Thanks!!