This report presents a meta-regression analysis exploring whether age and sex (percentage of male participants) explain variability in effect sizes derived from the provided imaging studies. The effect sizes are correlations related to GSH measures. Given the data available, we can only include mean age and percent male as moderators. Other potential moderators such as medication status, duration of psychosis, or symptom severity are not consistently reported and thus cannot be analyzed at this time.
We begin by loading the necessary packages and preparing the data.
The imaging_data frame and moderators data
frame are constructed based on the previously provided data.
# Imaging data as provided:
imaging_data <- data.frame(
authors = c(
# Positive (5)
"Matsuzawa et al. 2008", "Reyes-Madrigal et al. 2019",
"Iwata et al. 2021", "Coughlin et al. 2021", "Lesh et al. 2021",
# Negative (6)
"Matsuzawa et al. 2008", "Reyes-Madrigal et al. 2019",
"Iwata et al. 2021", "Coughlin et al. 2021", "Lesh et al. 2021", "Ravanfar et al. 2022",
# General (2)
"Reyes-Madrigal et al. 2019", "Iwata et al. 2021",
# Total (5)
"Matsuzawa et al. 2008", "Reyes-Madrigal et al. 2019",
"Iwata et al. 2021", "Lesh et al. 2021", "Ravanfar et al. 2022"
),
correlation = c(
# Positive
-0.43, 0.96, -0.08, 0.14, -0.266,
# Negative
-0.60, 0.36, 0.15, 0.21, -0.01, -0.348,
# General
0.14, -0.15,
# Total
-0.41, 0.42, -0.08, -0.293, -0.286
),
sample_size = c(
# Positive
20, 10, 67, 16, 33,
# Negative
20, 10, 67, 16, 33, 12,
# General
10, 67,
# Total
20, 10, 67, 33, 12
),
category = c(
rep("Positive", 5),
rep("Negative", 6),
rep("General", 2),
rep("Total", 5)
),
stringsAsFactors = FALSE
)
# Moderators data extracted from text (age and sex for a subset of imaging studies)
moderators <- data.frame(
authors = c("Matsuzawa et al. 2008", "Reyes-Madrigal et al. 2019",
"Iwata et al. 2021", "Coughlin et al. 2021", "Lesh et al. 2021"),
mean_age = c(30.7, 22.3, 43.6, 34.2, 21.4),
percent_male = c(0.60, 0.50, 0.75, 0.7391, 0.6944),
# Other moderators are NA due to lack of data
medication_status = c(NA, NA, "treated", "treated", NA),
duration_psychosis = c(NA, NA, NA, NA, NA),
overall_severity = c(NA, NA, NA, NA, NA),
positive_symptoms = c(NA, NA, NA, NA, NA),
negative_symptoms = c(NA, NA, NA, NA, NA),
disorganization_symptoms = c(NA, NA, NA, NA, NA),
stringsAsFactors = FALSE
)Next, we merge the effect size data (imaging_data) with
the moderator variables (moderators).
meta_imaging <- merge(imaging_data, moderators, by = "authors", all.x = TRUE)
# Subset to rows that have both mean_age and percent_male data
meta_imaging_sub <- meta_imaging[!is.na(meta_imaging$mean_age) & !is.na(meta_imaging$percent_male), ]
# Display the first few rows
head(meta_imaging_sub)## authors correlation sample_size category mean_age percent_male
## 1 Coughlin et al. 2021 0.21 16 Negative 34.2 0.7391
## 2 Coughlin et al. 2021 0.14 16 Positive 34.2 0.7391
## 3 Iwata et al. 2021 -0.15 67 General 43.6 0.7500
## 4 Iwata et al. 2021 0.15 67 Negative 43.6 0.7500
## 5 Iwata et al. 2021 -0.08 67 Total 43.6 0.7500
## 6 Iwata et al. 2021 -0.08 67 Positive 43.6 0.7500
## medication_status duration_psychosis overall_severity positive_symptoms
## 1 treated NA NA NA
## 2 treated NA NA NA
## 3 treated NA NA NA
## 4 treated NA NA NA
## 5 treated NA NA NA
## 6 treated NA NA NA
## negative_symptoms disorganization_symptoms
## 1 NA NA
## 2 NA NA
## 3 NA NA
## 4 NA NA
## 5 NA NA
## 6 NA NAWe convert the correlations to Fisher’s z-scores for meta-analysis
and compute their variances using the approximation
vi = 1/(n - 3):
With effect sizes (yi), their variances (vi), and moderators
(mean_age and percent_male), we can fit a random-effects meta-regression
model using the rma() function from
metafor.
res <- rma(yi = yi, vi = vi, mods = ~ mean_age + percent_male, data = meta_imaging_sub, method = "REML")
summary(res)##
## Mixed-Effects Model (k = 16; tau^2 estimator: REML)
##
## logLik deviance AIC BIC AICc
## -10.6185 21.2369 29.2369 31.4967 34.2369
##
## tau^2 (estimated amount of residual heterogeneity): 0.1965 (SE = 0.0992)
## tau (square root of estimated tau^2 value): 0.4433
## I^2 (residual heterogeneity / unaccounted variability): 84.09%
## H^2 (unaccounted variability / sampling variability): 6.29
## R^2 (amount of heterogeneity accounted for): 0.00%
##
## Test for Residual Heterogeneity:
## QE(df = 13) = 49.5832, p-val < .0001
##
## Test of Moderators (coefficients 2:3):
## QM(df = 2) = 2.0478, p-val = 0.3592
##
## Model Results:
##
## estimate se zval pval ci.lb ci.ub
## intrcpt 1.3042 0.9173 1.4218 0.1551 -0.4936 3.1020
## mean_age 0.0045 0.0186 0.2433 0.8078 -0.0319 0.0410
## percent_male -2.1762 1.7867 -1.2180 0.2232 -5.6780 1.3256
##
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1The output provides several pieces of information:
Conclusion:
The meta-regression model shows that neither mean age nor the proportion
of male participants explains the observed heterogeneity in effect
sizes. Although we attempted to explore these two moderators, the large
residual heterogeneity and non-significant results indicate that other
factors or variables—currently unmeasured or unavailable—are likely
driving the differences in effect sizes across studies.
To further reduce heterogeneity and better understand what drives these effect size differences, future analyses would need:
With more complete data, meta-regression could be extended to include these additional moderators and potentially reveal meaningful explanations for the observed heterogeneity.