In the target article Borsboom et al. argue convincingly that mental diseases are (mostly) not brain disorders, but represent highly complex network relations that depend on cultural, historical, and environmental mechanisms. They are highly variable across settings, across individuals, and even for the same individual in different settings, circumstances, and life periods. This narrative has major implications for the treatment and prevention of these conditions.

The new narrative explains why focusing on neurobiology (e.g., neurochemistry) to explain mechanisms and to develop effective treatments for mental conditions has achieved limited progress. The failure is so prominent that big pharma has largely abandoned new drug development in this field, despite its huge burden of disease and potential market (Chandler Reference Chandler2013). The largest meta-analyses to date show that for most mental health diseases, available drug treatments result in modest average treatment effects (d = 0.2–0.4) (Cipriani et al. Reference Cipriani, Furukawa, Salanti, Chaimani, Atkinson, Ogawa, Leucht, Ruhe, Turner, Higgins, Egger, Takeshima, Hayasaka, Imai, Shinohara, Tajika, Ioannidis and Geddes2018; Huhn et al. Reference Huhn, Tardy, Spineli, Kissling, Förstl, Pitschel-Walz, Leucht, Samara, Dold, Davis and Leucht2014; Leucht et al. Reference Leucht, Leucht, Huhn, Chaimani, Mavridis, Helfer, Samara, Rabaioli, Bächer, Cipriani, Geddes, Salanti and Davis2017), with small, incremental benefits over placebo. True treatment effects may be even smaller, if we consider biases (Ioannidis Reference Ioannidis2008). Some scientists even argue that extremely widely used drugs such as antidepressants are entirely ineffective and cause more harm than good (Gotzsche Reference Gotzsche2013), although this is probably an extreme position. For example, the recent largest meta-analysis on antidepressants found that almost all antidepressants were better than placebo for moderate/severe major depression, but the summary effect size for efficacy on a continuous scale was d = 0.30. There was also novelty bias: In head-to-head comparison trials, antidepressants seemed to work better when they were first marketed but then seemingly lost in efficacy as they became older (Cipriani et al. Reference Cipriani, Furukawa, Salanti, Chaimani, Atkinson, Ogawa, Leucht, Ruhe, Turner, Higgins, Egger, Takeshima, Hayasaka, Imai, Shinohara, Tajika, Ioannidis and Geddes2018). This further erodes the credibility of the estimated treatment effects in placebo-controlled trials, since these trials are mostly performed early in the licensing process when expectations are heightened.

Responses to drug and psychological therapies also show large between-person variability. Few patients have excellent responses, a modest proportion achieves some response, and many have no response. There is enormous investment in basic neuroscience research and intensive searches for informative biomarkers of treatment response and toxicity. The yield is close to nil. Even optimists acknowledge that, currently, there is still no clinically useful way to predict which patients will respond best to widely used medications such as antidepressants (Thase Reference Thase2014). If mental health problems are mostly not brain disorders, the dearth of useful neuroscience-derived biomarkers is only to be expected.

To overcome this dead end, we should shift emphasis away from the research paradigm that considers mental health problems to be mostly brain disorders and move towards exploring other, potentially more fruitful paths. First, this would mean reducing emphasis on identifying etiological brain pathways, and through them, biological markers and surrogate outcomes. If consistently strong and clinically useful biological markers/surrogates do not exist, perpetually searching for them would be in vain.

Second, the design of clinical trials in the field needs to be radically recast. Instead of running thousands of small trials of short-term duration and short-term response assessments, we could focus on larger simple trials with long-term follow-up (Ioannidis Reference Ioannidis2008). These trials should use a completely different core of non-biological, social outcomes likely to have relevance for most individuals. Such endpoints include suicides (completed/attempts), loss of job, marital and social relationships, social disability, personal finances (e.g., bankruptcy), and major quality of life and patient-related outcomes (Macefield Reference Macefield, Jacobs, Korfage, Nicklin, Whistance, Brookes, Sprangers and Blazeby2014).

Third, while such major endpoints are likely to be quite important to everyone, there are many other outcomes that are highly personalized. These personalized outcomes have to be defined for and by each patient, capturing what matters most under the specific personal and social context. One may discuss and choose before any treatment (either in an experimental trial setting or in real life) what context-specific outcomes have highest value in each case. What matters most may vary a lot across patients and may even change over time for the same patient, as life priorities and values evolve. Admittedly, evidence is weak to date on whether routine use of patient-reported outcome measures for feedback during the course of treatment improves the outcomes of mental disease (Kendrick et al. Reference Kendrick, El-Gohary, Stuart, Gilbody, Churchill, Aiken, Bhattacharya, Gimson, Brütt, de Jong and Moore2016). Simply sharing some information between patients and physicians may not suffice. Full personalized choice of the outcomes that matter may be needed, as has been described for other diseases, for example, substance use and chronic obstructive pulmonary disease (Alves et al. Reference Alves, Sales and Ashworth2017; Braid et al. Reference Braid, Baiardini, Molinengo, Garuti, Ferrari, Mantero, Blasi and Canonica2016).

Fourth, we could focus more on research for therapeutic and preventive interventions that have non-biological speculated mechanisms. In particular, we could prioritize understanding and ameliorating environmental and social stressors (Radua et al. Reference Radua, Ramella-Cravaro, Ioannidis, Reichenberg, Phiphopthatsanee, Amir, Yenn Thoo, Oliver, Davies, Morgan, McGuire, Murray and Fusar-Poli2018). Many of these may be context-specific, and they may vary across different cultures, times, epochs, and civilizations.

Fifth, some interventions that might be effective may need to be applied and delivered at a group or community level, or to the entire society, while others may need to be tailored to single patients. This means that we need to develop expanded research agendas both for community-level/society-level interventions and for personalized interventions. The most appropriate study designs would be different, depending on the goal of these trials. Cluster randomized trials may be most appropriate for group-level interventions. Personalized options may include adaptive trials (to account for strategies of sequential choices in long-term follow-up as response and/or treatment goals change) and n-of-1 trials.

Finally, much of the mental-health–related burden of disease may be induced or prevented by decisions in areas that have nothing to do with the brain, and go beyond the traditional remit of biomedical science. Our societies may need to consider more seriously the potential impact on mental health outcomes when making labor, education, financial and other social/political decisions at the workplace, state, country, and global levels. Mental health should be part of the conversation when different opinions exist on which decisions are preferable. Evidence on the impact of contesting actions could inform these decisions. Instead of thinking of mental disease as a narrow problem of brain tissue, brain cells, and brain molecules, we may need to think of it as an evolving, ever-changing challenge for society at large.