Below is a **publication-ready, JAMA Psychiatry–level methodological and results framework** tailored to your ketamine vs. placebo TRD meta-analysis. I explicitly separate **continuous (Hedges’ g)** and **binary (log OR / RR)** analytic streams where required. --- # 📐 STAGE 1 — EFFECT SIZE SELECTION & CALCULATION ## 1. Recommended Effect Sizes ### Continuous outcomes (primary): **Hedges’ g (bias-corrected standardized mean difference)** **Justification:** * Different depression scales (MADRS, HAMD-17, BDI-II) → requires standardization * Small sample sizes typical in ketamine RCTs → bias correction needed * Allows pooling across heterogeneous instruments ### Binary outcomes (secondary): **Log Odds Ratio (log OR)** (primary), optionally **Risk Ratio (RR)** **Justification:** * Response/remission are dichotomous * OR preferred in meta-analysis due to mathematical symmetry and compatibility with logistic models (Higgins et al., 2023) --- ## 2. Formulae ### A. Hedges’ g **Step 1: Raw mean difference** [ d = \frac{\bar{X}*{ketamine} - \bar{X}*{placebo}}{S_p} ] **Step 2: Pooled SD** [ S_p = \sqrt{\frac{(n_k - 1)S_k^2 + (n_p - 1)S_p^2}{n_k + n_p - 2}} ] **Step 3: Small-sample correction** [ g = J \cdot d, \quad J = 1 - \frac{3}{4(df) - 1} ] **Step 4: Variance** [ Var(g) = \frac{n_k + n_p}{n_k n_p} + \frac{g^2}{2(n_k + n_p - 2)} ] --- ### B. Log Odds Ratio [ OR = \frac{(a/b)}{(c/d)} = \frac{ad}{bc} ] [ \log(OR) = \ln\left(\frac{ad}{bc}\right) ] [ Var(\log OR) = \frac{1}{a} + \frac{1}{b} + \frac{1}{c} + \frac{1}{d} ] --- ## 3. Worked Example ### Continuous (Hedges’ g) * Ketamine: Mean = −15, SD = 8, n = 40 * Placebo: Mean = −5, SD = 7, n = 40 [ S_p = \sqrt{\frac{39(64) + 39(49)}{78}} = \sqrt{56.5} \approx 7.52 ] [ d = \frac{-15 - (-5)}{7.52} = \frac{-10}{7.52} = -1.33 ] [ J = 1 - \frac{3}{4(78)-1} \approx 0.990 ] [ g = -1.33 \times 0.990 = -1.32 ] --- ### Binary (Response) | | Response | No Response | | -------- | -------- | ----------- | | Ketamine | 24 | 16 | | Placebo | 10 | 30 | [ OR = \frac{24 \times 30}{16 \times 10} = 4.5 ] [ \log OR = \ln(4.5) = 1.50 ] --- ## 4. Conversion Formulae * OR → d: [ d = \frac{\ln(OR) \cdot \sqrt{3}}{\pi} ] * r → d: [ d = \frac{2r}{\sqrt{1 - r^2}} ] * t → d: [ d = \frac{t \sqrt{1/n_1 + 1/n_2}}{} ] --- ## 5. Interpretation Benchmarks * Cohen (1988): * Small = 0.2 * Medium = 0.5 * Large = 0.8 **Psychiatry-specific note:** Antidepressant meta-analyses often consider **g ≈ 0.3–0.5 clinically meaningful** (Cipriani et al., 2018). --- # 🌲 STAGE 2 — FOREST PLOT INTERPRETATION ## Structure * Each row = one RCT * Square = point estimate * Line = 95% CI * Vertical line = null (g = 0 or log OR = 0) * Diamond = pooled effect ## Individual Study * Position relative to null → direction of effect * CI width → precision * Box size → statistical weight ## Pooled Effect * Center = pooled estimate * Width = pooled CI * If CI excludes null → statistically significant ## Visual Heterogeneity * Tight clustering → low heterogeneity * Wide dispersion → high heterogeneity ## Subgroups * Separate panels (e.g., single vs. repeated dosing) * Compare subgroup diamonds --- ## 📄 Ready-to-paste Results paragraph "A total of [k] randomized controlled trials were included in the meta-analysis. Individual study effect sizes are presented in the forest plot (Figure X). Most studies demonstrated an effect favoring ketamine over placebo, with varying degrees of precision as indicated by the width of confidence intervals. The pooled effect estimate (Hedges’ g = [VALUE], 95% CI [LOWER, UPPER]) was statistically significant, as the confidence interval did not cross the null value. Visual inspection suggested [low/moderate/high] heterogeneity, with [degree of overlap] among individual study confidence intervals." --- # 📊 STAGE 3 — HETEROGENEITY ## Q Statistic [ Q = \sum w_i (y_i - \bar{y})^2 ] * df = k − 1 * Null: homogeneity **Reporting:** "Q([df]) = [VALUE], p = [VALUE]" ⚠️ Limitation: underpowered when k small --- ## I² [ I^2 = \frac{Q - df}{Q} \times 100% ] Interpretation (Higgins et al., 2003): * 25% = low * 50% = moderate * 75% = high ⚠️ Relative measure (depends on precision) --- ## Tau² and Prediction Interval [ PI = \bar{y} \pm 1.96\sqrt{\tau^2} ] Template: "Despite statistical heterogeneity (I² = [X]%), the prediction interval suggests that in 95% of similar populations the true effect would range from [LOWER] to [UPPER]." --- ## Model Selection * Fixed: assumes one true effect * Random: distribution of effects ✔ Use **random-effects (REML or Paule-Mandel)** ✔ Apply **HKSJ correction** (IntHout et al., 2014) --- ## Exploratory Analyses * Subgroup → Q_between * Meta-regression → coefficient = change in effect per unit moderator * Leave-one-out → influence diagnostics --- # 🔍 STAGE 4 — PUBLICATION BIAS ## Funnel Plot * X-axis: effect size * Y-axis: SE Symmetry → no bias Asymmetry → bias or heterogeneity --- ## Egger’s Test [ Z_i = a + b \cdot \frac{1}{SE_i} ] Template: "Egger’s test indicated [no/significant] asymmetry (intercept = [X], SE = [Y], t = [Z], p = [W])." --- ## Trim-and-Fill Report: * Observed pooled estimate * Adjusted estimate * Imputed studies (k) ⚠️ Adjusts asymmetry, not true bias --- ## Fail-Safe N (Orwin) "An additional [N] null studies would be required to reduce the effect size to g = 0.20." --- ## GRADE * Funnel asymmetry → downgrade evidence * Integrate with RoB 2 --- # 📝 STAGE 5 — NARRATIVE SYNTHESIS ## A. Opening Sentence "A meta-analysis of [k] studies (N = [total participants]) demonstrated a [statistically significant] [large] effect of ketamine versus placebo on depressive symptom severity in TRD (Hedges’ g = [VALUE], 95% CI [LOWER, UPPER], p = [VALUE])." --- ## B. Heterogeneity "Heterogeneity was [low/moderate/high] (Q([df]) = [VALUE], p = [VALUE]; I² = [VALUE]%). The estimated between-study variance was τ² = [VALUE], corresponding to τ = [VALUE]. The prediction interval ranged from [LOWER] to [UPPER], indicating that the true effect in similar clinical settings may vary substantially." --- ## C. Subgroup Findings "Subgroup analyses indicated that [moderator] significantly influenced treatment effects (Q_between = [VALUE], p = [VALUE]). Specifically, [group A] demonstrated a larger effect (g = [VALUE]) compared to [group B] (g = [VALUE])." --- ## D. Publication Bias "Visual inspection of the funnel plot suggested [symmetry/asymmetry]. Egger’s test indicated [no/significant] asymmetry (p = [VALUE]). Trim-and-fill analysis imputed [k] studies, yielding an adjusted effect size of [VALUE], which [did/did not] materially differ from the observed estimate. Orwin’s fail-safe N indicated that [N] null studies would be required to attenuate the effect to a trivial magnitude." --- ## E. Clinical Significance * NNT: [ NNT = \frac{1}{CER \cdot (RR - 1)} ] * U3: [ U_3 = \Phi(g) ] Example statement: "This effect corresponds to approximately [X]% of patients in the ketamine group outperforming the average placebo patient (Cohen’s U3). The estimated number needed to treat (NNT) for response was [VALUE], indicating that one additional patient benefits for every [VALUE] treated. On the MADRS scale, this corresponds to an absolute reduction of approximately [VALUE] points." --- ## F. Limitations * Protocol heterogeneity * Small k * Unblinding due to dissociation * Selective reporting * GRADE: [low/moderate/high] --- ## G. Conclusion "Ketamine demonstrates a rapid and clinically meaningful antidepressant effect in treatment-resistant depression. However, substantial heterogeneity and methodological limitations temper confidence in the magnitude and generalizability of effects. Future trials should focus on optimizing dosing strategies, evaluating maintenance protocols, and conducting head-to-head comparisons with established treatments." --- # ⚠️ FINAL NOTE This is a **dual-outcome meta-analysis**: * Continuous → Hedges’ g (primary efficacy signal) * Binary → log OR / RR (clinical translation) 👉 These must be analyzed **separately** and only integrated narratively. --- If you want next step, I can: * Generate **R (metafor) or Stata code** * Help compute **your actual dataset** * Draft a **full JAMA-style Methods section**