Meta-analyses are an incredibly useful tool for synthesising evidence. However, such analyses typically use aggregate data, meaning the average scores or outcomes for treatment groups, which can cause problems if we’re trying to dig a little deeper into the question of ‘what works’ to answer ‘what works, and for whom?’
The ‘for whom?’ question is especially important in terms of implications for practice. Doctors and patients faced with a choice of treatment might be concerned that reviews only provide summary findings, telling us about the average treatment effect for a whole population. Patients might want to know if the treatment is more or less likely to work for them, taking into account their personal characteristics (such as age and gender) and their specific illness profile (for example, how severe their depression is or how long they’ve been suffering from it.) In fact, we can use trials to answer these questions, by looking at whether there is a moderator effect on the outcome.
A moderator is a variable that impacts on the strength of an effect – in this case, whether a treatment is effective or not. The easiest way to think of a moderator (although a little dull!) is as a light dimmer switch. The switch itself is a particular treatment or intervention, and the light is the effect of that treatment – in this case, an improvement in symptoms of depression. In terms of our ‘what works, for whom?’ question, we can think of most trials as evaluating simply whether the switch works. Does flicking the switch (giving someone the treatment) actually turn on the light (reduce symptoms of depression)? Moderator analyses go a step further to ask whether any factors act as a dimmer, making the light brighter or weaker, i.e. making the treatment more or less effective.
However, the problem with checking for these kinds of effects in typical meta-analyses is that using only the summary statistics from the individual trials can be misleading. Using the individual patient data from each trial is a way around this, and this is exactly what Nelson and colleagues have done in order to look at moderators of antidepressant treatment in older people.
In an IPD (individual patient data) analysis, the researchers conducting the meta-analysis ask the authors of each trial in the review to provide their original data sets for the analysis. This means that rather than having to take average scores from the trials, the authors are able to combine the individual data from each patient in each trial into one overall data set. This gives them much greater statistical power to detect effects, and also avoids a problem called ‘the ecological fallacy’ (whereby aggregate data can obscure moderator effects).
The authors had conducted a previous review of trials comparing antidepressants to placebo in patients aged 60 or above, who had major depressive disorder and were living in the community. They identified 10 trials, and were able to access the data on all 10. Seven of those trials contained the information they were looking for on moderators (specifically, they looked at age, age of onset, sex, single episode vs recurrent depression, initial depression severity and cognitive impairment). The 7 trials included a total of 2,488 patients of whom 1,494 received antidepressants and 994 received a placebo.
Unsurprisingly, overall they found that antidepressants were more effective than placebo
- Turning to the question of moderator effects:
- Course of illness (single episode or recurrent depression), sex, and age were not significantly related to the treatment group-response interaction
- Duration of illness was found to be significant (having a significant interaction with treatment group and depression outcome)
- Severity was also a significant moderator, but only for those patients with an illness duration of more than 10 years
- In both cases, the effect of antidepressants (compared to placebo) was greater in those groups.
The authors concluded:
Our findings suggest that the duration of depressive illness moderates antidepressant response in patients over age 60. Patients with an illness duration >10 years showed greater drug-placebo differences, while those with a short illness duration showed no drug effect.
This suggests that not only is the brightness turned down when there has been short illness duration, but you may as well have turned the light off. The authors report no effect of antidepressants for those patients who had depression for less than 2 years. Adding in severity to those with a longer illness duration further narrows down the subgroup of patients who showed benefit, with the authors noting that this subgroup would only comprise 385 patients. Although we can’t be certain that the equivalent numbers would be found in a typical primary care population, this would suggest that in fact only a minority of older people with depression show benefit from antidepressants. As the authors state, “Clinicians should reconsider the prevailing practice of simply prescribing an antidepressant in these patients”.
For those with a long illness duration and at least moderate severity however, antidepressants have a beneficial effect, demonstrating how analyses such as these can help us better tailor treatment recommendations to specific patients who will show most benefit.
The authors were not able to look at cognitive status as a moderator due to limitations in how the data was reported, with some studies only reporting the threshold rather than the actual scores. This demonstrates that even with IPD analyses where reviewers are accessing the ‘raw’ data, inconsistencies in the way that measures are used or findings reported can still disrupt our ability to synthesise data
- Similarly, the authors comment that assessing illness duration was complicated as trials didn’t provide detail on how this was assessed. They suggest the significance of illness duration in this analysis indicates that duration should be more reliably and routinely collected in future trials
- Finally, it should be noted that there was large effect of the study covariate on response (meaning effect size varied a lot between studies), suggesting unexplained heterogeneity in the included studies. Although this was controlled for in the analyses, it might indicate that important factors underlying variation were missed out.
Nelson JC, Delucchi KL, Schneider LS. Moderators of outcome in late-life depression: a patient-level meta-analysis. Am J Psychiatry. 2013 Jun 1;170(6):651-9. doi: 10.1176/appi.ajp.2012.12070927. [PubMed abstract]
Riley RD, Lambert PC, Abo-Zaid G. Meta-analysis of individual participant data: rationale, conduct, and reporting. BMJ. 2010 Feb 5;340(feb05 1):c221–c221.
Lambert PC, Sutton AJ, Abrams KR, Jones DR. A comparison of summary patient-level covariates in meta-regression with individual patient data meta-analysis. J Clin Epidemiol. 2002 Jan;55(1):86–94. [PubMed abstract]
Moderators of outcome in late-life depression: should we be prescribing antidepressants to older people?: Meta… http://t.co/TVW1hRiK1j
@dr_know helps answer this question: Are antidepressants effective for elderly patients with depression? http://t.co/zFQ2WwRdy0
My new blog for @Mental_Elf: Are antidepressants effective for elderly patients with depression? http://t.co/kv2SX4EkCV
Pls RT @UKCochraneCentr New patient level meta-analysis questions prescribing antidepressants to older people http://t.co/zFQ2WwRdy0
Pls RT @CLAHRC_CP New patient level meta-analysis questions prescribing antidepressants to older people http://t.co/zFQ2WwRdy0
Pls RT @DementiaUK New patient level meta-analysis questions prescribing antidepressants to older people http://t.co/zFQ2WwRdy0
Pls RT @ncmh_wales New patient level meta-analysis questions prescribing antidepressants to older people http://t.co/zFQ2WwRdy0
Pls RT @ageukcampaigns New patient level meta-analysis questions prescribing antidepressants to older people http://t.co/zFQ2WwRdy0
@Mental_Elf @ageukcampaigns Should be questioned for all ages.
Pls RT @MindCharity New patient level meta-analysis questions prescribing antidepressants to older people http://t.co/zFQ2WwRdy0
Antidepressants vary in effectiveness depending on how long patients have been depressed, says new meta-analysis http://t.co/zFQ2WwRdy0
Antidepresivos sólo serían eficaces para personas que llevan más de 2 años deprimidas, según nuevo estudio
New blog on treatment moderators, and why individual patient data meta-analysis helps us look for them http://t.co/kv2SX4EkCV #statistics
Antidepressants may not be effective for elderly patients who have had depression for less than 2 years http://t.co/zFQ2WwRdy0
What can individual patient data analyses tell us that normal meta analyses don’t? http://t.co/zFQ2WwRdy0 #Depression #LaterLife
How does understanding moderators help us tailor treatment decisions for older people with depression? http://t.co/zFQ2WwRdy0
@Mental_Elf Quire. How to choose with the person in front of you. There’s the challenge for each n of 1
RT @Mental_Elf: Antidepressants most effective for people over 60 if they have had depression for more than 10 years http://t.co/zFQ2WwRdy0
What works and for whom? Should we be prescribing antidepressants to older people? -http://t.co/055wsjCMNm
Don’t miss: Moderators of outcome in late-life depression: should we be prescribing antidepressants to older people? http://t.co/zFQ2WwRdy0
@Mental_Elf If a professional thinks it may help – I don’t see why not. Especially if other ideas have been tried and failed.
Mental Elf: Moderators of outcome in late-life depression: should we be prescribing antidepressants to older people? http://t.co/AFo2UjKlyy
@sadneurons Duration of illness and severity of illness were both found to be significant moderators http://t.co/zFQ2WwRdy0
@sadneurons Course of illness, sex and age were not significantly related to the treatment group-response interaction http://t.co/zFQ2WwRdy0
In case you missed it on your bank holiday – http://t.co/ugvh3nXa7C my @Mental_Elf post on moderators of antidepressant tx outcomes
@dr_know Bank holiday or not, your blog got a great response :-) http://t.co/zFQ2WwRdy0 #Depression #Antidepressants #Moderators #LaterLife
If anyone has a more interesting metaphor for moderators than a dimmer switch btw, let me know. Asking for a friend. http://t.co/ugvh3nXa7C
Runners up: @IanCummins9 http://t.co/TKN2fUvkX5 @field_matt http://t.co/BM7qt3Q6WB @dr_know http://t.co/zFQ2WwRdy0
[…] data sets in greater detail than we can do when using summary results. For an overview, I described here the benefits of what are known as ‘individual patient data meta-analyses’ for answering […]