Festschrift_Ellen.qmd

---
title: "Fundamental Flaws in the Appraisal of Adolescent Depression Treatment"
subtitle: "In honour of Dr Ellen Leibenluft"
author: "Professor Argyris Stringaris, MD, PhD, FRCPsych"
format: revealjs
logo: ucl_brain_sciences.png
fontsize: "22pt"
include-in-header:
  - text: |
      <style>
      .reveal .slide-logo {
        max-height: unset;
        height: 150px;
      }
      </style>
---

## Overview

-   Importance of Adolescent Depression

-   How Do We Treat Adolescent Depression?

-   The Problem With Comparing Treatments

-   Vast Differences in the Control Conditions between Treatment Types.

-   Dubious Control Groups for Psychotherapy

-   Understanding Human Expectations and Psychiatric Treatments

## What I learnt from Ellen

![](ArgyrisandEllen.png)

## What I learnt from Ellen

Ellen,

-   gave meaning to the concept of irritability and more generally that of negative and aggressive **mood**, something with vast public health implications.

-   injected rationality into an extraordinary debate that of the US "Pediatric Bipolar", and curbing excesses that came from very powerful places.

-   is consistently a **Clinical Researcher** with an instinct for what matters to patients and public.

-   worked together with an extraordinary team of people and bringing out the most of each.

-   is an extraordinary writer.

-   has a keen sense and knowledge of history and therefore human fallibility.

## Importance of Adolescent Depression

![](adolescent_mental_health.jpeg){fig-align="left"}

## How Do We Treat Adolescent Depression?

![National Institute of Clinical Excellence Guidelines](NICE_1.png){fig-align="left"}

## How Do We Treat Adolescent Depression

![National Institute of Clinical Excellence Guidelines](NICE_2.png){fig-align="left"}

## How Do We Treat Adolescent Depression

![American Academy of C&A Psychiatry Practice Parameter](AACAP_1.png){fig-align="left"}

## How do We Treat Adolescent Depression

![](AACAP_2.png){fig-align="left"}

## Comparing Treatments

These recommendations are not based on head to head trials between modalities.

-   only one head-to-head trial exists in adolescent depression.
-   comparisons rely on the comparison of the results of randomised controlled trials (RCTs) of each modality against RCTs of the other.
-   In other words, they rest on comparison of comparisons.

## Comparing Treatments: Clinical Importance

This has potentially vast implications for the health of the public:

-   As a patient/carer: which treatment should I/my child take?

-   As a health professional: you can take A or/over B

-   As a public health official: doctors should ...

## The Problem With Comparing Treatments

-   In antidepressant studies, the control is the placebo pill

-   In psychotherapy studies controls vary substantially

    What is tested is:

$$Effect_{Med} = Effect_{Psy}$$

where *Effect* denotes the effect on depression that either medication (*Med* ) or psychotherapy (*Psy*) have.

## The Problem With Comparing Treatments

In particular, it relies on the equality of controls:

$$Effect_{MedActive} - Effect_{MedControls} = Effect_{PsyActive} - Effect_{PsyControls}$$

or:

$$ Effect_{MedControls} = Effect_{PsyControls}$$

## The Problem with Comparing Treatments

![](appendicitis.png){fig-align="left" width="340"}

Compare: antibiotics vs surgery:

-   Antibiotics improvement rate: 40% on active; 2% on control

-   Surgery improvement rate: 80% on active, 42% on control

## The Problem with Comparing Treatments

Compare: antibiotics vs surgery:

-   Antibiotics improvement rate: 40% on active; 0% on control

-   Surgery improvement rate: 80% on active, 40% on control

Would you tell your patient that these are equally good treatments?

## Comparing Treatments: Mechanistic Implications

If there is variability in control arms, then:

-   is it random?

-   does it tell us something about how treatments work? e.g. :

    -   what are the psychological mechanisms of placebo?

    -   why would people find other control groups unfavourable?

-   more generally, what is an appropriate counterfactual to treatment?

## Assessing Treatment Comparisons: Methods

-   Extracted baseline and follow up means and sds for each study (where available) from all randomised controlled trials (RCTs) in adolescent depression using medication or psychotherapy.

-   Estimated Standardised Mean Differences (SMD) for each study.

    $$SMD_{change} = \frac{Mean_{t_{2}} - Mean_{t_{1}}}{\frac{SD_{t{2}} + SD_{t{1}}}2}\  $$

## Assessing Treatment Comparisons: Methods

We also needed the estimation of the standard errors. For this we require the test-retest correlation coefficient of the instrument, which is not reported in studies:

$$SE_{change} = \sqrt{\frac{r_{t_{1}t_{2}}}{n} + \frac{SMD^2}{2n}}$$

We therefore simulated 1000 datasets with correlations coefficients from a broad distribution and used these for our calculations–the magnitude of $r_{t_{1}t_{2}}$ made very little difference to results.

## Assessing Treatment Comparisons: Methods

Random Effects Meta-Regression

$$Y_i = x_i\beta_i + u_i + \epsilon_i$$

where

$$ \begin{aligned}
    Υ_i &=
    \begin{cases}
        0 & MedControl:  b_0 + u_i + \epsilon_i\\
        1 & MedActive:   b_0 + b_{1_{i}} +  u_i + \epsilon_i \\
        2 & PsyActive:   b_0 + b_{2_{i}}  +  u_i + \epsilon_i \\
        3 & PsyControl:  b_0 + b_{3_{i}}  +  u_i + \epsilon_i \\
    \end{cases}
\end{aligned}$$

## Assessing Treatment Comparisons: Results					

There were 81 studies in total

-   antidepressants: 28 active arms and 25 control arms of antidepressant trials;

-   psychotherapy: 55 active arms and 52 control arms from psychotherapy trials.

-   Placebo was the control condition for all medication trials

-   WL controls, care as usual and several other conditions such as attention control for psychotherapy

## Assessing Treatment Comparisons: SMDs overall

![](plot_means_all.png){fig-align="left"}

## Assessing Treatment Comparisons: SMDs CBT and SRI only

![](plot_means_cbt.png){fig-align="left"}

## Assessing Treatment Comparisons: SMDs no waitlist

![](plot_means_wl.png){fig-align="left"}

## Assessing Treatment Comparisons: differences

![](coef_pic.png){fig-align="left" width="361"}

## What were the controls? 

-   Mostly waitlist (WL) and care-as-usual (CAU).

-   Others were attention controls, which were typically poorly matched to active groups.

-   Often information was missing on key variables.

## What is a good control? 

Controls should serve as counterfactuals: what would an individual look like if they did not receive treatment?

There should be a **latent true value** of what an individual would look like had they not received the treatment.

## What is a good control? 

A good control is crucially **ΝΟΤ** being told that you were allocated to:

-   simply waiting (WL control)

-   getting what they would have gotten anyway (CAU control), which is heterogeneous (and in our field sometimes not much at all).

-   receiving a sub-par treatment that the experimenter themselves do not believe in (some ill-designed control).

    All this also whilst **knowing** that someone else (other people in the trial) got the "real thing". This creates negative counterfactual compared to treatment.

## Are there alternative explanations for the finding? 

-   could it be due to different instruments used?

    -   No, even when using common instruments, such as CDRS and HAM-D, the results were the same.

-   could there be more regression to the mean in the medication group.

    -   There could be, the values are higher at baseline in most studies (with common instruments)

    -   But the difference is very large, even when matching studies with the same baseline values.

    -   We are thinking of ways of dealing with this statistically at present.

## Where to next?

At a practical level:

-   change guidelines, at the very least they should be toned down.

-   impose more rigorous controls on psychotherapy studies.

-   actively research the social and neurobiological underpinnings of control conditions. This has to include:

    -   expectancy effects

    -   opportunity cost

-   think about the ethical implications of control groups.