g_lineplot function - connect non-continuous points

[wwojciech]

Original message

g_lineplot does not connect points for non-continuous data points.

Example to reproduce

adlb <- scda::synthetic_cdisc_data("latest")$adlb
adlb <- dplyr::filter(
  adlb, 
  ANL01FL == "Y", 
  PARAMCD == "ALT", 
  AVISIT != "SCREENING",
  !(ARM == "A: Drug X" & AVISIT == "WEEK 2 DAY 15")
) %>%
droplevels()

g_lineplot(adlb, subtitle = "Laboratory Test: ALT")

It would be great to assess whether g_lineplot could be extended so that such non-continuous points are connected.
See also MNG01 in STREAM, where non-continuous data points are connected.

TODO

Use Xiaoying's example code, and connect non-continious points, see below
#149 (comment)

library(ggplot2)

set.seed(4321)

df = data.frame(
  mean = ifelse(rbinom(10, 1, 0.8) == 1, runif(10, 0, 15), NA),
  upper = ifelse(rbinom(10, 1, 0.8) == 1, runif(10, 15, 20), NA),
  lower = ifelse(rbinom(10, 1, 0.8) == 1, runif(10, -5, 0), NA),
  ARM = rep(c('A', 'B'), each = 5), 
  VISIT = rep(1:5, 2)
) 

pd = position_dodge(0.8)

df %>% 
  filter(!is.na(mean)) %>% 
  mutate(upper = ifelse(is.na(lower), NA, upper),
         lower = ifelse(is.na(upper), NA, lower)) %>% 
  ggplot(aes(VISIT, mean, color = ARM)) +
  theme_bw() +
  geom_point(position = pd) +
  geom_line(position = pd) + 
  geom_errorbar(aes(ymin = lower, ymax = upper), position = pd) 

[yangx61]

Consider removing NA in df_stats before apply ggplot. See example below

library(ggplot2)

set.seed(4321)

df = data.frame(
  AVAL = ifelse(rbinom(10, 1, 0.8) == 1, runif(10, 0, 15), NA),
  ARM = rep(c('A', 'B'), each = 5), 
  VISIT = rep(1:5, 2)
) 

ggplot(df, aes(VISIT, AVAL, color = ARM)) +
  theme_bw() +
  geom_point() +
  geom_line()

ggplot(na.omit(df), aes(VISIT, AVAL, color = ARM)) +
  theme_bw() +
  geom_point() +
  geom_line()

[wwojciech]

@yangx61 The issue is a bit more involved here, as df_stats (supplied to data argument of the ggplot function) is in wide format rather than long, and as you know na.omit removes record(s) with at least one NA column. Hence you cannot simply use na.omit(df_stats) as you may remove rows with non-NA values for some series.

By wide format I mean the series (columns) with stats itself, not the ARM, AVISIT of course.

[shajoezhu]

Interesting! @wwojciech
Can you give an example please?

[wwojciech]

The example would be a non-NA midpoint (e.g. mean) column and e.g. lower CI column with NA, both for the same record.

[yangx61]

If there is NA in one of the CI values, then the error bars should not be displayed. So we can just remove NA from the mean. See example below.

library(ggplot2)

set.seed(4321)

df = data.frame(
  mean = ifelse(rbinom(10, 1, 0.8) == 1, runif(10, 0, 15), NA),
  upper = ifelse(rbinom(10, 1, 0.8) == 1, runif(10, 15, 20), NA),
  lower = ifelse(rbinom(10, 1, 0.8) == 1, runif(10, -5, 0), NA),
  ARM = rep(c('A', 'B'), each = 5), 
  VISIT = rep(1:5, 2)
) 

pd = position_dodge(0.8)
ggplot(df, aes(VISIT, mean, color = ARM)) +
  theme_bw() +
  geom_point(position = pd) +
  geom_line(position = pd) + 
  geom_errorbar(aes(ymin = lower, ymax = upper), position = pd)

df %>% 
  filter(!is.na(mean)) %>% 
  mutate(upper = ifelse(is.na(lower), NA, upper),
         lower = ifelse(is.na(upper), NA, lower)) %>% 
  ggplot(aes(VISIT, mean, color = ARM)) +
  theme_bw() +
  geom_point(position = pd) +
  geom_line(position = pd) + 
  geom_errorbar(aes(ymin = lower, ymax = upper), position = pd)