Effectiveness of Psychotherapies and Psychotropic Drugs against Mental Illness/Psychotherapies for Depression

Effectiveness of Psychotherapies and Psychotropic Drugs against Mental Illness Psychotherapies for depression Psychotherapies for anxiety Loading libraries library(dplyr) library(metafor) library(meta) Part 1: Data At https://docs.metapsy.org/databases/depression-psyctr/ you can download documents in PDF format and data in ZIP format related to a meta-analysis conducted by researchers entitled "Depression: Psychotherapy vs. Control," which analyzed 500 clinical trials of psychotherapies for depression, comparing groups undergoing psychotherapy and control groups. In the csv dataset, for each study there is the Hedges effect size g with its relative standard deviation, the variable recruitment which allows to catalogue the studies in clin (institutional carried out in hospitals or clinics) or com (population recruited through advertisements or media), the variable condition_arm1 which contains the type of psychotherapy used in the study: (cbt= Cognitive-Behavioral Therapy, bat= Behavioral Activation Therapy, pst= Problem solving Therapy, ipt= Interpersonal Psychotherapy, dyn= Short-term Psychodynamic Psychotherapy, lrt= life review therapy, other psy= other type of psychotherapy), the variable condition_arm2 which contains the type of group with which the group undergoing psychotherapy is compared: (cau= Care-as-usual: The standard psychosocial or medical support that a patient normally receives in hospital or clinic, wl= Waiting List: Where patients do not receive any immediate treatment, pha= Pharmacological treatment, other ctr= other type of inactive control group, sup= supportive counseling, other psy= other type of psychotherapy). To calculate the odds ratio OR I use the 4 variables event_arm1,event_arm2,totaln_arm1,totaln_arm2 which are respectively: Number of events (responders, remission, deterioration cases) in the first trial arm, Number of events (responders, remission, deterioration cases) in the second trial arm, Total number of participants arm 1 (event data), Total number of participants arm 2 (event data). Loading and viewing data: df <- read.csv("data.csv", sep = ";", dec = ",", stringsAsFactors = FALSE) df %>% filter(!is.na(event_arm1)) %>% select(study,.g,.g_se,recruitment,condition_arm1,condition_arm2,event_arm1,event_arm2,totaln_arm1,totaln_arm2) %>% head() study .g .g_se recruitment condition_arm1 condition_arm2 event_arm1 event_arm2 totaln_arm1 totaln_arm2 Alexopoulos, 2016 -0.215183490629242 0.174866416963007 oth pst other ctr 28 37 84 87 Araya, 2021 0.265099475177847 0.0811923132832991 clin bat cau 159 114 440 440 Araya, 2021 0.395725691602631 0.109735001466973 clin bat cau 108 70 217 215 Barber, 2012 0.0989335546931013 0.25150795223941 com dyn other ctr 14 12 51 50 Barrett, 2001 0.270056935486176 0.260130882861446 clin pst other ctr 21 16 37 36 Beutel, 2014 0.519091790686381 0.194425538658389 oth dyn cau 34 18 78 78 Part 2: Meta-analysis Calculating Hedges' g We calculate Hedges' g in meta-analyses of studies with recruitment (populations recruited through advertisements or the media) and perform the Egger test to assess whether there is publication bias due to the fact that studies with positive results and high g are more likely to be published than those with modest or negative results. Since the p-value is <0.05, there is indeed publication bias in the test, so we use Duval and Tweedie's "trim and fill" procedure, which estimates the number of missing studies and recalculates g, obtaining g = 0.53 with a 95% confidence interval: [ 0.46 , 0.6 ], so in this case the effect of psychotherapy on depression is medium. # Filter for studies in community settings df_com <- df %>% filter(recruitment=="com") # Random-effects meta-analysis with Hartung-Knapp modification m_g <- metagen(TE = .g, seTE = .g_se, data = df_com, studlab = study, random = TRUE, method.tau = "DL", # DerSimonian-Laird method.random.ci = "HK") # Modifica Hartung-Knapp # Egger test for publication bias metabias(m_g, method.stat = "linreg") Linear regression test of funnel plot asymmetry Test result: t = 7.80, df = 443, p-value < 0.0001 Bias estimate: 1.6125 (SE = 0.2068) Details: - multiplicative residual heterogeneity variance (tau^2 = 3.8474) - predictor: standard error - weight: inverse variance - reference: Egger et al. (1997), BMJ m_g<- trimfill(m_g) print(paste("g =",round(m_g$TE.random,2),"; 95%CI: [",round(m_g$lower.random,2),",",round(m_g$upper.random,2),"]")) "g = 0.53; 95%CI: [ 0.46 , 0.6 ]" Calculating g in meta-analyses for studies with recruitment=clin (institutional studies conducted in hospitals or clinics) yields g=0.32, so in this case the effect of psychotherapy on depression is small: # Filter for studies in institutional settings df_inst <- df %>% filter(recruitment=="clin") # Random-Effects Meta-Analysis with Hartung-Knapp Modification m_g <- metagen(TE = .g, seTE = .g_se, data = df_inst, studlab = study, random = TRUE, method.tau = "DL", # DerSimonian-Laird method.random.ci = "HK") # Modifica Hartung-Knapp m_g <- trimfill(m_g) print(paste("g =",round(m_g$TE.random,2),"; 95%CI: [",round(m_g$lower.random,2),",",round(m_g$upper.random,2),"]")) "g = 0.32; 95%CI: [ 0.18 , 0.45 ]" Measuring g for the various psychotherapies immediately after treatment (time = post), we obtain that lrt= life review therapy has the highest g (g=0.81), bat= Behavioral Activation Therapy (g=0.49), cbt= Cognitive Behavioral Therapy (g=0.47) etc. , so the effect of most psychotherapies on depression is medium... for (p in unique(df$condition_arm1)) { df_post <- df %>% filter(time == "post", condition_arm1==p) m_pst<-metagen(TE = .g, seTE = .g_se, data = df_post, studlab = study, random = TRUE, # Modello a effetti casuali method.tau = "DL", # DerSimonian-Laird method.random.ci = "HK") # Modifica Hartung-Knapp m_pst<- trimfill(m_pst) print(paste(p,": g=",round(m_pst$TE.random,2),"; 95%CI: [",round(m_pst$lower.random,2),",",round(m_pst$upper.random,2),"]")) } [1] "cbt : g= 0.47 ; 95%CI: [ 0.38 , 0.55 ]" [1] "pst : g= 0.37 ; 95%CI: [ 0.1 , 0.65 ]" [1] "lrt : g= 0.81 ; 95%CI: [ 0.38 , 1.24 ]" [1] "3rd : g= 0.55 ; 95%CI: [ 0.36 , 0.74 ]" [1] "other psy : g= 0.48 ; 95%CI: [ 0.37 , 0.59 ]" [1] "bat : g= 0.49 ; 95%CI: [ 0.36 , 0.63 ]" [1] "sup : g= 0.37 ; 95%CI: [ 0.08 , 0.65 ]" [1] "dyn : g= 0.35 ; 95%CI: [ 0.07 , 0.62 ]" [1] "ipt : g= 0.25 ; 95%CI: [ 0.1 , 0.4 ]" Calculation of OR (Odds Ratio) Measuring the OR by comparing the psychotherapy groups with the control groups (waiting list, usual care, or placebo), we obtain OR = 1.82, i.e., the probability of a positive response to psychotherapy for depression increases by +82% compared to control groups: df_bin <- df %>% filter(!is.na(event_arm1), !is.na(event_arm2), !is.na(totaln_arm1), !is.na(totaln_arm2)) es_bin <- escalc( measure = "OR", ai = event_arm1, bi = totaln_arm1 - event_arm1, ci = event_arm2, di = totaln_arm2 - event_arm2, data = df_bin ) # Meta-analysis on the calculated results (random effects model) res_psy_ctr <- rma( yi, vi, data = es_bin %>% filter( condition_arm2 %in% c("wl", "cau", "pla", "other ctr")), method = "DL" # Method DerSimonian-Laird ) res_corretto <- trimfill(res_psy_ctr) print(paste("OR =", round(exp(res_corretto$b),2))) [1] "OR = 1.82"