Abstract

Let us begin by asking: How good are our current treatments for depression in adolescents? As recently reviewed by Emslie in this journal (Emslie 2012), over the past two decades there has been an enormous increase in the number of pharmacological and psychological treatment studies for adolescent depression; however, we clearly have a long way to go. Overall remission rates in the Treatment for Adolescents with Depression Study (TADS) (Kennard, et al. 2006) and Treatment of Resistant Depression in Adolescents (TORDIA) (Vitiello, et al. 2011) were 50% or less, while overall response rates for pharmacological and psychosocial treatments tend to hover at approximately 60%. In other words, the cup is a bit more than half-full but there is clearly considerable room to do even better.
New Pharmacological Treatments
Much has been written about the thinness of the pipeline for psychopharmacology across the lifespan. In part this may be due to a decrease in pharmaceutical industry research funding for new treatment compounds (Nutt and Goodwin 2011); this may be explained by numerous factors including: the long time necessary to develop psychopharmacological agents; the high failure rate in the development of medicines to treat psychiatric disorders (especially when compared to other disorders); and changes in expected return on investment from these new compounds when they are finally marketed. While there are ongoing Phase II studies of compounds with new mechanisms of action (e.g. NK1, NMDA, CRF, AMPA, mGluR5, and mGluR2) that may be useful in depression, almost none have yet reached Phase III trials. In any event, the future will likely bring breakthroughs (some small and some larger), however, over the next five years significant improvements in our treatment of adolescent depression are unlikely to come from new pharmacologic agents. With this in mind, we can either despair or consider what else we have learned from clinical trials and try to apply these to clinical management of our patients.
Adding Together Many Small Process Improvements
It is common to find large differences in treatment outcome from site to site, even in pharmacological treatment studies (Spirito et al. 2009). This is a clear indication that factors other than our “core” treatments have a major impact on outcomes. Further, it is likely that these differences are not a single, big factor but, rather, the sum of many small effects. In other words, becoming a “good” therapist with above average patient outcomes isn't likely to be the result of learning one particular treatment skill or method; instead, it is more likely to be a function of knowing and using many skills.
While the comparison may seem superficial, improving treatment of adolescent depression is not dissimilar to today's greater automotive mile per gallon ratings that have come from dozens of small improvements (e.g., reduced friction, variable value control, turning the engine off during brief stops, torque converter lockup, improved tires, weight reduction, improved lubricants, etc.) and not from a fewer number of medium or large improvements (Decicco and Ross 1996). Similarly, for adolescent depression, we must consider the possibility that just a dozen small process improvements in our treatments for adolescent depression, which each improve outcomes by a mere 2% on average, will reduce the number of medication non-responders by 50%! Some obvious candidates for such changes include: reducing barriers to attending sessions; better motivation for the adolescent to engage in treatment; building a better liaison with both parents; treating current psychiatric disorders in other family members; etc. However, a moment's reflection makes it clear that we are still trying to address this problem at a level that is too high. For example, there are many potential barriers to attending treatment sessions, and they will require many different solutions, including providing services in schools and churches; more conveniently locating provider offices; finding incentives for trained clinicians to work in rural settings; and using televideo for those youth living in isolated communities and to reach those for whom it is impractical, unsafe, or otherwise impossible to travel to clinical services. With this in mind, consideration of small process improvements may be out best chance of improving the care of our patients, at least in the near term.
If we make a few straightforward assumptions, small, incremental process improvements can be relatively straightforward to implement, and even study. First, it is reasonable to assume that most process improvements (in both pharmacological and psychotherapeutic treatment) that actually make treatment better are likely to be helpful across psychiatric disorders, across treatment modalities, and across service delivery sites. To arrive at this conclusion, we don't need to assume that all process improvements are equally good; we must just be confident that they go in the same direction for all cases (i.e., they don't help in some circumstances and hurt in others). Second, we must assume that we don't have significant interaction terms that wipe out the good effect of successful process improvements; it is “OK” if two improvements together give you somewhat more or less than the sum of the parts but we should be confident that, when taken together, two good process improvements don't yield worse results. With these assumptions in place, we can readily screen putative process improvements, one-at-a-time, in a general adolescent psychiatric population, keeping all the ones that give better outcomes and even combining them.
Finally, we need to gather cost data so that we can get some rank ordering of “bang-for-the-buck.” While randomized controlled studies are the gold standard for therapeutic efficacy, econometric approaches to large-sample, naturalistic data can also be of value. Our studies, whether controlled or naturalistic, must also acquire cost data—small improvements added together only work if, individually, they are also relatively economical. And, especially in individual practices, small process improvements must also make economic sense.
While we await the studies that will address new treatments and treatment processes, we can begin to look for these opportunities in our own practices. Two-percent gains here-and-there in treatment processes can quickly add up. So, while we await new medications or psychotherapies, individual practitioners can and should examine the processes of their treatments to find those modest but important gains by improving attendance, adherence, support systems, and the other “little things” that truly make a difference.
Personalized Medicine
“My object all sublime, I shall achieve in time…”
(Gilbert, Sullivan 1885)
Cancer treatment provides great examples of the value of personalized medicine. Many of the treatments are expensive. Most have significant side effects. Well worked-out, basic science gives strong a priori reasons why certain treatments will be effective only in a subgroup of patients with a particular cancer—e.g., treatment depends on the presence of a particular receptor or target a single genetic mutation that is sometimes but not always found with that malignancy. In psychiatry, we are just beginning to link genetic changes with response to a particular treatment (Novarino et al. 2012) and, for now, these are likely to be found in autism where we have the best evidence for specific genes playing a significant role in a specific psychiatric disorder. Further progress in this area requires a deeper genetic understanding than we have at present for disorders such as depression, anxiety, ADHD, or schizophrenia. But, we should not be discouraged as new data are quickly being assembled, and this will change in the not-too-distant future.
In spite of this current situation in psychiatry, are there other opportunities to personalize treatments for “our” disorders? First, we discuss a model that conveniently ignores questions about patient preference, cost, and side effects (we will address them later). We will also address simply the question of initial treatment but the same principles will apply to optimizing treatment algorithms. With these simplifications, the question in basic form reduces to whether or not we can separate patients into two groups: In one group treatment A is most effective and in the other, treatment B is most effective. But, we must be careful to note that some relationships between predictors and treatment response do not result in useful data to improve clinical decision-making for personalized medicine. These include: • Overall predictors of better or worse response may be of some value in letting us know that one has a higher or lower likelihood for overall response but if treatment A is better than treatment B for both groups then this doesn't change treatment. • A predictor of somewhat worse response for treatment A and better response for treatment B, but not of sufficient magnitude such that treatment B is better than treatment A, may be scientifically useful but it doesn't help us in selecting B as the treatment for anybody.
Now let's consider side effects and patient (and family) preferences. They can be handled one of two ways. • Youth, families, practitioners, and other stakeholders can help us develop a consensus utility function that integrates the likelihood of response to a specific treatment, while also accounting for the side effects expected from that treatment. This approach is more tractable but it ignores the fact that different children and families are likely to have a wide range in their evaluation of the trade-offs between treatment and side effects. • Provide response and side effect data to the patient and family, based on their individual characteristics, thus allowing them to make the ultimate decision in collaboration with the clinician.
Ultimately the first approach may be better for developing treatment algorithms and the second approach better to actually use with patients.
For our exemplar condition, unipolar depression in adolescents, do we have predictors which let us personalize treatment based on expected responses? One place the field has looked is in genes that are related to drug metabolism (Maalouf et al. 2011) but, as of now, these genes have not lead to different treatment choices. However, history of abuse is associated with poorer response to treatment across several studies of depression in youth; but, abuse worsens treatment outcomes to both pharmacological and psychotherapeutic approaches, so this cannot lead to changes in treatment approach (Lewis et al. 2010; Shamseddeen et al. 2011). Nonetheless, there are still great opportunities for personalized medicine approaches to be explored by targeting groups with child and adolescent depression who have known poor response. A good example is those youth whose parents are currently depressed and/or youth in families with high conflict. For each, specific interventions are available and can be applied with relatively simple, small process changes in care.
In the future, single genes, single environmental measures, and single subject characteristics may provide us with personalized approaches to treating some psychiatric disorders. And, eventually, we may even be able to combine genes, environmental measures, imaging measures, etc. While this is an area of active investigation, that time has not yet arrived, Research groups are applying modern machine learning approaches that provide efficient learning of non-linear functions, e.g., support vector machines to large sets of potential response correlates (Orru et al. 2012). These algorithms, which Google uses to drive driverless cars, are currently being used in development of predictive biomarkers across medicine (Han et al. 2012), and are being explored to map functional connectivity data to the diagnosis of ADHD in youth (Eloyan et al. 2012). But, this work is in its earliest stages and will not be as readily available in the near future. Small process changes are available to us now.
Discussion
This is but one approach to changing treatment outcomes in the present, at both a group and individual clinician level. Fortunately, there are other potential models for improving our treatments. These will be discussed in a subsequent article.
Ultimately improvement in our treatments will come from better basic understanding of the underlying pathophysiology, better individualized treatment, and better treatment processes. None will be easy and attention to all is necessary to achieve the best possible results. In the meantime, while we wait for the data to arrive, focusing on small process changes in our treatments can make a difference in our patients' care, today.
