Skip to contents

Stratified Analysis

Source: vignettes/stratified-analysis.Rmd
stratified-analysis.Rmd

Stratified regression repeats the analysis inside each subgroup and places the results side by side. It is useful when the same association may look different across groups.

library(gtregression)
library(dplyr)

data("data_birthwt", package = "gtregression")

birthwt_data <- data_birthwt |>
  mutate(
    race = factor(race, levels = c(1, 2, 3),
                  labels = c("White", "Black", "Other")),
    smoke = factor(smoke, levels = c(0, 1), labels = c("No", "Yes")),
    ht = factor(ht, levels = c(0, 1), labels = c("No", "Yes")),
    ui = factor(ui, levels = c(0, 1), labels = c("No", "Yes")),
    low = factor(low, levels = c(0, 1), labels = c("Normal BW", "Low BW")),
    ptl_cat = factor(ifelse(ptl > 0, "Yes", "No"), levels = c("No", "Yes"))
  )

attr(birthwt_data$age, "label") <- "Maternal age"
attr(birthwt_data$lwt, "label") <- "Maternal weight"
attr(birthwt_data$smoke, "label") <- "Smoking during pregnancy"
attr(birthwt_data$ht, "label") <- "Hypertension"
attr(birthwt_data$ui, "label") <- "Uterine irritability"
attr(birthwt_data$ptl_cat, "label") <- "Previous preterm labour"

Describe by Stratum

Start with a descriptive table by the stratifying variable. This is the companion table for the stratified regression: it helps users see the size and clinical profile of each subgroup before fitting stratum-specific models.

strata_desc <- descriptive_table(
  data = birthwt_data,
  exposures = c("age", "lwt", "smoke", "ht", "ui", "ptl_cat"),
  by = race,
  percent = column,
  show_overall = last,
  theme = clinical
)

strata_desc$table

Characteristic

White, N=96

Black, N=26

Other, N=67

Overall, N=189

Maternal age

23.5 (20.0-29.0)

20.5 (17.2-24.0)

22.0 (19.0-25.0)

23.0 (19.0-26.0)

Maternal weight

129.5 (112.0-143.2)

129.0 (120.0-179.0)

119.0 (105.0-130.0)

121.0 (110.0-140.0)

Smoking during pregnancy

No

44 (45.8%)

16 (61.5%)

55 (82.1%)

115 (60.8%)

Yes

52 (54.2%)

10 (38.5%)

12 (17.9%)

74 (39.2%)

Hypertension

No

91 (94.8%)

23 (88.5%)

63 (94.0%)

177 (93.7%)

Yes

5 (5.2%)

3 (11.5%)

4 (6.0%)

12 (6.3%)

Uterine irritability

No

83 (86.5%)

23 (88.5%)

55 (82.1%)

161 (85.2%)

Yes

13 (13.5%)

3 (11.5%)

12 (17.9%)

28 (14.8%)

Previous preterm labour

No

82 (85.4%)

22 (84.6%)

55 (82.1%)

159 (84.1%)

Yes

14 (14.6%)

4 (15.4%)

12 (17.9%)

30 (15.9%)

Categorical variables shown as n (%); percentages are by column.

Continuous variables shown as Median (IQR).

Univariable by Stratum

stratified_uni_reg() fits one model per exposure inside each stratum. The result is a single wide table, with one spanner per stratum.

strata_uni <- stratified_uni_reg(
  data = birthwt_data,
  outcome = low,
  exposures = c("age", "lwt", "smoke", "ht", "ui", "ptl_cat"),
  stratifier = race,
  approach = logit,
  theme = clinical
)

strata_uni$table

race = White

race = Black

race = Other

Characteristic

N

OR (95% CI)

p-value

N

OR (95% CI)

p-value

N

OR (95% CI)

p-value

Maternal age

96

0.95 (0.86–1.04)

0.226

26

1.05 (0.90–1.23)

0.526

67

0.94 (0.84–1.05)

0.297

Maternal weight

96

0.98 (0.97–1.00)

0.123

26

0.99 (0.97–1.01)

0.517

67

0.97 (0.95–1.00)

0.056

Smoking during pregnancy

96

26

67

No

Ref.

Ref.

Ref.

Yes

5.76 (1.78–18.60)

0.003

3.30 (0.63–17.16)

0.156

1.25 (0.35–4.46)

0.731

Hypertension

96

26

67

No

Ref.

Ref.

Ref.

Yes

2.22 (0.35–14.20)

0.399

3.11 (0.24–39.54)

0.382

5.59 (0.55–56.99)

0.146

Uterine irritability

96

26

67

No

Ref.

Ref.

Ref.

Yes

2.26 (0.66–7.75)

0.196

3.11 (0.24–39.54)

0.382

2.88 (0.80–10.33)

0.105

Previous preterm labour

96

26

67

No

Ref.

Ref.

Ref.

Yes

5.96 (1.80–19.72)

0.003

1.44 (0.17–12.23)

0.736

4.47 (1.18–16.90)

0.027

Abbreviations: OR = Odds Ratio; CI = Confidence Interval.

Ref. = reference category.

Full Multivariable Model by Stratum

With adjust_for = NULL, stratified_multi_reg() fits one multivariable model inside each stratum using all supplied exposures.

strata_full <- stratified_multi_reg(
  data = birthwt_data,
  outcome = low,
  exposures = c("age", "lwt", "smoke", "ht", "ui", "ptl_cat"),
  stratifier = race,
  approach = logit,
  theme = clinical
)

strata_full$table

race = White

race = Black

race = Other

Characteristic

N

Adjusted OR (95% CI)

p-value

N

Adjusted OR (95% CI)

p-value

N

Adjusted OR (95% CI)

p-value

Maternal age

96

0.97 (0.86–1.08)

0.548

26

0.87 (0.64–1.19)

0.391

67

0.93 (0.81–1.07)

0.305

Maternal weight

96

0.99 (0.97–1.01)

0.333

26

0.97 (0.94–1.01)

0.136

67

0.97 (0.94–1.00)

0.074

Smoking during pregnancy

96

26

67

No

Ref.

Ref.

Ref.

Yes

3.35 (0.94–12.02)

0.063

16.50 (0.91–298.21)

0.058

0.81 (0.17–3.85)

0.794

Hypertension

96

26

67

No

Ref.

Ref.

Ref.

Yes

3.43 (0.39–30.08)

0.265

85.06 (0.60–11,959.18)

0.078

6.71 (0.52–86.08)

0.143

Uterine irritability

96

26

67

No

Ref.

Ref.

Ref.

Yes

1.02 (0.22–4.73)

0.978

67.61 (1.42–3,225.31)

0.033

2.60 (0.65–10.43)

0.176

Previous preterm labour

96

26

67

No

Ref.

Ref.

Ref.

Yes

4.68 (1.17–18.72)

0.029

4.87 (0.11–208.59)

0.409

4.13 (0.91–18.77)

0.066

Abbreviations: OR = Odds Ratio; CI = Confidence Interval.

Ref. = reference category.

Complete observations included by race stratum: White: N = 96; Black: N = 26; Other: N = 67

Exposure-Specific Adjusted Models by Stratum

Use adjust_for when each exposure should be adjusted for the same variables within each stratum. This mirrors multi_reg(adjust_for = ...), but repeats the same workflow separately inside each stratum.

strata_multi <- stratified_multi_reg(
  data = birthwt_data,
  outcome = low,
  exposures = c("smoke", "ht", "ui", "ptl_cat"),
  stratifier = race,
  adjust_for = c("age", "lwt"),
  approach = logit,
  theme = striped
)

strata_multi$table

race = White

race = Black

race = Other

Characteristic

N

Adjusted OR (95% CI)

p-value

N

Adjusted OR (95% CI)

p-value

N

Adjusted OR (95% CI)

p-value

Smoking during pregnancy

96

26

67

No

Ref.

Ref.

Ref.

Yes

4.97 (1.47–16.80)

0.010

2.96 (0.48–18.32)

0.243

1.23 (0.32–4.80)

0.762

Hypertension

96

26

67

No

Ref.

Ref.

Ref.

Yes

3.72 (0.45–30.67)

0.222

5.71 (0.27–121.78)

0.264

7.93 (0.66–95.10)

0.102

Uterine irritability

96

26

67

No

Ref.

Ref.

Ref.

Yes

1.59 (0.43–5.96)

0.488

4.49 (0.28–72.28)

0.289

2.68 (0.72–10.05)

0.143

Previous preterm labour

96

26

67

No

Ref.

Ref.

Ref.

Yes

6.26 (1.77–22.16)

0.004

0.96 (0.09–9.85)

0.973

5.55 (1.31–23.56)

0.020

Abbreviations: OR = Odds Ratio; CI = Confidence Interval.

Ref. = reference category.

Adjusted for age and lwt

Complete observations included by race stratum: White: N = 96; Black: N = 96; Other: N = 96; White: N = 96; Black: N = 26; Other: N = 26; White: N = 26; Black: N = 26; Other: N = 67; White: N = 67; Black: N = 67; Other: N = 67

If a stratum cannot fit a model, the function skips that stratum with a warning and continues. This is intentional: sparse strata are common in real data, and one small subgroup should not erase the whole analysis.

Forest Plot by Stratum

forest_df() can also prepare one stratified regression object for forest_reg(). The variable rows are kept once and each stratum is placed in a side-by-side effect column, which is easier to compare than repeating the full variable list for every subgroup.

strata_forest_data <- forest_df(strata_multi)

forest_reg(
  strata_forest_data,
  ci_col_width = 18
)

What To Inspect

  • $table: rendered side-by-side table.
  • $table_display: wide data used to build the table.
  • $per_stratum: full per-stratum result objects.
  • $models: fitted models by stratum.
  • $model_summaries: summaries for the fitted models.
  • $variable_labels: display labels used in the wide table.
  • $reg_check: diagnostics for linear models.