2021 Aravena Systematic Review
Packages
pacman::p_load(tidyverse,
meta,
gtsummary,
janitor)theme_set(theme_minimal())Dataset and cleaning
Load the data
df <- read_csv("2021_aravena_systematic_review_data_sinus.csv")##
## ── Column specification ────────────────────────────────────────────────────────
## cols(
## .default = col_character(),
## mean_c = col_double(),
## sd_c = col_double(),
## n_c = col_double(),
## mean_e = col_double(),
## sd_e = col_double(),
## n_e = col_double(),
## `Riss_Bias(0High_7Low)` = col_double(),
## event.c = col_double(),
## n.c = col_double(),
## event.e = col_double(),
## n.e = col_double()
## )
## ℹ Use `spec()` for the full column specifications.df <- df %>%
janitor::clean_names() %>%
rename(risk_bias = riss_bias_0high_7low) Extract the year
df <- df %>%
mutate(year = as.numeric(str_extract(author_year, "[0-9]+"))) %>%
relocate(year, .after = author_year)Create two datasets: cont and bin
Dataset continuous
df_cont <- df %>%
select(author_year:n_e,
year,
risk_bias)Dataset binary
df_bin <- df %>%
select(author_year,
time,
event_c:n_e_2,
year,
risk_bias)EDA
df %>%
pivot_longer(c(n_c, n_e),
names_to = "sample",
values_to = "sample_values") %>%
mutate(sample = recode(sample, n_c = "Control", n_e = "Intervention")) %>%
ggplot(aes(x = sample_values)) +
geom_histogram(bins = 5) +
facet_grid(sample ~ .) +
labs(title = "Sample size per groups, Cont",
y = "Count",
x = "Sample size")
df %>%
pivot_longer(c(n_c_2, n_e_2),
names_to = "sample",
values_to = "sample_values") %>%
mutate(sample = recode(sample, n_c_2 = "Control", n_e_2 = "Intervention")) %>%
ggplot(aes(x = sample_values)) +
geom_histogram(bins = 5) +
facet_grid(sample ~ .) +
facet_grid(sample ~ .) +
labs(title = "Sample size per groups, Bin",
y = "Count",
x = "Sample size")
Continuous metanalysis
mean difference (argument sm = “MD”)
m_cont <- meta::metacont(n_e,
mean_e,
sd_e,
n_c,
mean_c,
sd_c,
studlab = author_year,
data = df_cont,
sm = "SMD")m_cont## SMD 95%-CI %W(fixed) %W(random)
## Borges F., et al. (2011) 0.1239 [-0.5925; 0.8404] 15.9 13.7
## Fouad W., et al. (2018) 5.6721 [ 3.5236; 7.8206] 1.8 8.4
## Khaled H., et al. (2019) 1.5935 [ 0.5581; 2.6288] 7.6 12.6
## Trinh H., et al. (2019) 5.1838 [ 3.5744; 6.7932] 3.1 10.4
## Millán A., et al. (2020) 0.3334 [-0.2309; 0.8977] 25.6 14.2
## Ranaan J., et al. (2018) 1.3817 [ 0.6654; 2.0981] 15.9 13.7
## Nedir R., et al. (2017) 0.8603 [-0.0912; 1.8118] 9.0 12.9
## Quian S., et al. (2020) -0.0527 [-0.6734; 0.5680] 21.1 14.0
##
## Number of studies combined: k = 8
##
## SMD 95%-CI z p-value
## Fixed effect model 0.7750 [0.4895; 1.0604] 5.32 < 0.0001
## Random effects model 1.5713 [0.6351; 2.5076] 3.29 0.0010
##
## Quantifying heterogeneity:
## tau^2 = 1.5268 [1.1016; 13.4936]; tau = 1.2356 [1.0496; 3.6734];
## I^2 = 89.4% [81.6%; 94.0%]; H = 3.08 [2.33; 4.07]
##
## Test of heterogeneity:
## Q d.f. p-value
## 66.33 7 < 0.0001
##
## Details on meta-analytical method:
## - Inverse variance method
## - DerSimonian-Laird estimator for tau^2
## - Jackson method for confidence interval of tau^2 and tau
## - Hedges' g (bias corrected standardised mean difference)forest(m_cont,
leftcols = c("author_year", "n_c", "n_e") ,
leftlabs = c("Author (Year)", "n (control)", "n (experimental)"),
sortvar = TE,
comb.random = TRUE,
comb.fixed = FALSE,
test.overall.random = TRUE,
label.right = "Favours experimental",
label.left = "Favours control",
predict = TRUE)
Verificar que pasa con Trihn y Fuad
MA sin esos dos
df_cont %>%
filter(author_year != "Fouad W., et al. (2018)") %>%
filter(author_year != "Trinh H., et al. (2019)") %>%
# el ma
meta::metacont(n_e,
mean_e,
sd_e,
n_c,
mean_c,
sd_c,
studlab = author_year,
data = .,
sm = "SMD") ## SMD 95%-CI %W(fixed) %W(random)
## Borges F., et al. (2011) 0.1239 [-0.5925; 0.8404] 16.7 17.3
## Khaled H., et al. (2019) 1.5935 [ 0.5581; 2.6288] 8.0 12.8
## Millán A., et al. (2020) 0.3334 [-0.2309; 0.8977] 26.9 19.8
## Ranaan J., et al. (2018) 1.3817 [ 0.6654; 2.0981] 16.7 17.3
## Nedir R., et al. (2017) 0.8603 [-0.0912; 1.8118] 9.5 13.9
## Quian S., et al. (2020) -0.0527 [-0.6734; 0.5680] 22.2 18.9
##
## Number of studies combined: k = 6
##
## SMD 95%-CI z p-value
## Fixed effect model 0.5382 [0.2455; 0.8309] 3.60 0.0003
## Random effects model 0.6405 [0.1174; 1.1635] 2.40 0.0164
##
## Quantifying heterogeneity:
## tau^2 = 0.2772 [0.0224; 2.5556]; tau = 0.5265 [0.1497; 1.5986];
## I^2 = 66.7% [20.7%; 86.0%]; H = 1.73 [1.12; 2.68]
##
## Test of heterogeneity:
## Q d.f. p-value
## 15.03 5 0.0102
##
## Details on meta-analytical method:
## - Inverse variance method
## - DerSimonian-Laird estimator for tau^2
## - Jackson method for confidence interval of tau^2 and tau
## - Hedges' g (bias corrected standardised mean difference)Esos dos le dan toda la heterogeneidad, hay que explicar porque esos dos muestran que la intervención es efectiva, mientras que el resto no
Binary metanalysis
df_bin %>%
metabin(event_e,
n_e_2,
event_c,
n_c_2,
data = .,
sm = "OR",
method = "Inverse",
studlab = author_year)## OR 95%-CI %W(fixed) %W(random)
## Borges F., et al. (2011) 3.3529 [0.1307; 86.0329] 14.3 14.3
## Fouad W., et al. (2018) NA 0.0 0.0
## Khaled H., et al. (2019) NA 0.0 0.0
## Trinh H., et al. (2019) NA 0.0 0.0
## Millán A., et al. (2020) NA 0.0 0.0
## Ranaan J., et al. (2018) 1.1176 [0.1492; 8.3734] 37.1 37.1
## Nedir R., et al. (2017) 0.5625 [0.0465; 6.8057] 24.2 24.2
## Quian S., et al. (2020) 0.5278 [0.0440; 6.3372] 24.4 24.4
##
## Number of studies combined: k = 4
##
## OR 95%-CI z p-value
## Fixed effect model 0.9224 [0.2704; 3.1459] -0.13 0.8973
## Random effects model 0.9224 [0.2704; 3.1459] -0.13 0.8973
##
## Quantifying heterogeneity:
## tau^2 = 0 [0.0000; 7.0901]; tau = 0 [0.0000; 2.6627];
## I^2 = 0.0% [0.0%; 53.5%]; H = 1.00 [1.00; 1.47]
##
## Test of heterogeneity:
## Q d.f. p-value
## 0.99 3 0.8042
##
## Details on meta-analytical method:
## - Inverse variance method
## - DerSimonian-Laird estimator for tau^2
## - Jackson method for confidence interval of tau^2 and tau
## - Continuity correction of 0.5 in studies with zero cell frequenciesdf_bin %>%
metabin(event_e,
n_e_2,
event_c,
n_c_2,
data = .,
sm = "OR",
method = "Inverse",
studlab = author_year) %>%
forest(.,
leftcols = c("author_year", "n_c_2", "n_e_2") ,
leftlabs = c("Author (Year)", "n (control)", "n (experimental)"),
sortvar = TE,
comb.random = TRUE,
comb.fixed = FALSE,
test.overall.random = TRUE,
label.right = "Favours experimental",
label.left = "Favours control",
predict = TRUE)