Advertisement

From Computation to Clinic

Open AccessPublished:April 02, 2022DOI:https://doi.org/10.1016/j.bpsgos.2022.03.011

      Abstract

      Theory-driven and data-driven computational approaches to psychiatry have enormous potential for elucidating mechanism of disease and providing translational linkages between basic science findings and the clinic. These approaches have already demonstrated utility in providing clinically relevant understanding, primarily via back translation from clinic to computation, revealing how specific disorders or symptoms map onto specific computational processes. Nonetheless, forward translation, from computation to clinic, remains rare. In addition, consensus regarding specific barriers to forward translation—and on the best strategies to overcome these barriers—is limited. This perspective review brings together expert basic and computationally trained researchers and clinicians to 1) identify challenges specific to preclinical model systems and clinical translation of computational models of cognition and affect, and 2) discuss practical approaches to overcoming these challenges. In doing so, we highlight recent evidence for the ability of computational approaches to predict treatment responses in psychiatric disorders and discuss considerations for maximizing the clinical relevance of such models (e.g., via longitudinal testing) and the likelihood of stakeholder adoption (e.g., via cost-effectiveness analyses).

      Keywords

      Translation of research findings into clinical settings to solve clinical problems is a primary challenge of modern psychiatry (
      • Yip S.W.
      • Kiluk B.
      • Scheinost D.
      Toward addiction prediction: An overview of cross-validated predictive modeling findings and considerations for future neuroimaging research.
      ). This requires a coordinated effort between at least 3 constituencies: basic research (“bench”), clinical research (“bedside”), and the community (“stakeholders”) (
      • Cohrs R.J.
      • Martin T.
      • Ghahramani P.
      • Bidaut L.
      • Higgins P.J.
      • Shahzad A.
      Translational medicine definition by the European Society for Translational Medicine.
      ). Within this framework, clinical insights and experience motivate novel basic research, while novel basic research motivates novel therapeutic approaches.
      Computational psychiatry aims to use advances in computational cognitive neuroscience and machine learning to improve knowledge about mental health conditions and their treatment (
      • Huys Q.J.M.
      • Maia T.V.
      • Frank M.J.
      Computational psychiatry as a bridge from neuroscience to clinical applications.
      ) and, as such, is an intrinsically translational field. However, the direction of translation so far has mostly been one-directional: from clinic to computation. Computational approaches have been deployed in many ways to shed light on the cognitive and neurobiological structure of established psychiatric descriptions and classifications, but rarely to discover novel descriptions or create new interventions. Closing this translational loop by bringing these insights back into the clinic encounters numerous challenges, many of which are faced by the broader neuroscience field in general (
      • Huys Q.J.M.
      • Maia T.V.
      • Frank M.J.
      Computational psychiatry as a bridge from neuroscience to clinical applications.
      ). However, promising avenues for overcoming these challenges are now emerging using theory-driven, data-driven, and hybrid approaches. Here, we suggest new directions for research in this area, discuss challenges, and propose solutions to maximize translation from computation to clinic.

      From the Bottom Up: Toward Algorithmic Development of Novel Therapies

      Computational models find theoretical appeal in their ability to help bridge levels of abstraction when describing a neural system’s function, i.e., from the implementation level to the algorithmic level to the computational level (
      • Marr D.
      • Poggio T.
      From understanding computation to understanding neural circuitry: Artificial Intelligence Laboratory AI Memo 357.
      ). The implementational level describes how the neural system is set up [e.g., which neurons encode rewards or punishments, and how are they connected to other neurons (
      • Haber S.N.
      The place of dopamine in the cortico-basal ganglia circuit.
      )]. The algorithmic level describes, in mathematical terms, the way in which the input (e.g., experience with rewards and/or punishments) is transformed to an output (e.g., conditioned responding or instrumental behavior), as in reinforcement learning (
      • Schultz W.
      • Dayan P.
      • Montague P.R.
      A neural substrate of prediction and reward.
      ). The computational level describes what the system is seeking to achieve (e.g., obtaining nutrition or avoiding harm).
      As noted previously, this framework may be particularly well suited to understanding and refining existing psychiatric interventions (
      • Nair A.
      • Rutledge R.B.
      • Mason L.
      Under the hood: Using computational psychiatry to make psychological therapies more mechanism-focused.
      ). Many of the medical interventions at a clinician’s disposal have been discovered, at least in part, as a result of serendipity and may therefore lack a full-fledged mechanistic theory to account for their efficacy. A well-known example in psychiatry is the case of antipsychotic medications for schizophrenia: their proposed mechanism of action during initial development is different from what we now know (
      • Robinson E.
      Psychopharmacology: From serendipitous discoveries to rationale design, but what next?.
      ). It was subsequently hypothesized that their pharmacological action was largely dependent on their affinity for dopamine receptors (
      • Kapur S.
      • Mamo D.
      Half a century of antipsychotics and still a central role for dopamine D2 receptors.
      ). Models of dopamine that bridge implementation (i.e., dopamine neurons in the midbrain and their projection to the striatum and prefrontal cortex), algorithm (e.g., the temporal difference learning model), and computation (e.g., delusional beliefs, hallucinations, and/or apathy) are valuable for understanding and optimizing pharmacological interventions for schizophrenia (
      • Braver T.S.
      • Barch D.M.
      • Cohen J.D.
      Cognition and control in schizophrenia: A computational model of dopamine and prefrontal function.
      ,
      • Maia T.V.
      • Frank M.J.
      An integrative perspective on the role of dopamine in schizophrenia.
      ,
      • Smith A.J.
      • Li M.
      • Becker S.
      • Kapur S.
      Linking animal models of psychosis to computational models of dopamine function.
      ,
      • Heinz A.
      • Murray G.K.
      • Schlagenhauf F.
      • Sterzer P.
      • Grace A.A.
      • Waltz J.A.
      Towards a unifying cognitive, neurophysiological, and computational neuroscience account of schizophrenia.
      ,
      • Adams R.A.
      • Stephan K.E.
      • Brown H.R.
      • Frith C.D.
      • Friston K.J.
      The computational anatomy of psychosis.
      ) but were largely developed post hoc, after the initial demonstration of the efficacy of antipsychotic medications.
      However, this type of analysis implies that there may be potential for therapies to be developed and optimized at the algorithmic level alone, by focusing exclusively on the mathematical form of the psychological process that is thought to be dysfunctional in the disorder (Figure 1). We assume that a dysfunctional psychological process that is causally related to a given symptom can be described by an algorithm, which is built from a set of parameters that correspond to the core components of the process. By selectively altering algorithmic parameters or altering system inputs (e.g., providing more information) via clinical intervention, it is assumed that this will change the dysfunctional process in a way specified by the algorithm and ultimately result in an improvement of the clinical condition (
      • Nair A.
      • Rutledge R.B.
      • Mason L.
      Under the hood: Using computational psychiatry to make psychological therapies more mechanism-focused.
      ). An area in which such principles have already been adopted is within the reinforcer pathology framework for understanding substance use disorders (
      • Bickel W.K.
      • Johnson M.W.
      • Koffarnus M.N.
      • MacKillop J.
      • Murphy J.G.
      The behavioral economics of substance use disorders: Reinforcement pathologies and their repair.
      ). For example, substance users show a greater preference for immediate over delayed rewards (
      • Bickel W.K.
      • Johnson M.W.
      • Koffarnus M.N.
      • MacKillop J.
      • Murphy J.G.
      The behavioral economics of substance use disorders: Reinforcement pathologies and their repair.
      ), which can be described by algorithmic delay-discounting models (
      • Peters J.
      • Miedl S.F.
      • Büchel C.
      Formal comparison of dual-parameter temporal discounting models in controls and pathological gamblers.
      ). Behavioral manipulations [e.g., of episodic future thinking or working memory (
      • Stein J.S.
      • Wilson A.G.
      • Koffarnus M.N.
      • Daniel T.O.
      • Epstein L.H.
      • Bickel W.K.
      Unstuck in time: Episodic future thinking reduces delay discounting and cigarette smoking.
      ,
      • Bickel W.K.
      • Yi R.
      • Landes R.D.
      • Hill P.F.
      • Baxter C.
      Remember the future: Working memory training decreases delay discounting among stimulant addicts.
      )], which reduce the discounting of future rewards, can also reduce drug consumption and thus represent a causal pathway for modulating drug intake and a possible avenue for treatment. A related example is the effect of cost on demand: demand for cigarettes can be affected by their price, so increasing cost per cigarette suppresses demand (
      • González-Roz A.
      • Jackson J.
      • Murphy C.
      • Rohsenow D.J.
      • MacKillop J.
      Behavioral economic tobacco demand in relation to cigarette consumption and nicotine dependence: A meta-analysis of cross-sectional relationships.
      ). Algorithmic models of demand [e.g., (
      • Hursh S.R.
      • Silberberg A.
      Economic demand and essential value.
      )] might be applied to determine the level of tobacco taxation that optimally suppresses demand and increases revenue (
      • MacKillop J.
      • Few L.R.
      • Murphy J.G.
      • Wier L.M.
      • Acker J.
      • Murphy C.
      • et al.
      High-resolution behavioral economic analysis of cigarette demand to inform tax policy.
      ) or evaluate contingency management therapies, i.e., monetary reinforcement of successful abstinence (
      • González-Roz A.
      • García-Pérez Á.
      • Weidberg S.
      • Aonso-Diego G.
      • Secades-Villa R.
      Reinforcer pathology and response to contingency management for smoking cessation.
      ,
      • Yoon J.H.
      • Suchting R.
      • McKay S.A.
      • San Miguel G.G.
      • Vujanovic A.A.
      • Stotts A.L.
      • et al.
      Baseline cocaine demand predicts contingency management treatment outcomes for cocaine-use disorder.
      ,
      • Yoon J.H.
      • Suchting R.
      • de Dios C.
      • Vincent J.N.
      • McKay S.A.
      • Lane S.D.
      • Schmitz J.M.
      Decreased cocaine demand following contingency management treatment.
      ,
      • Regier P.S.
      • Redish A.D.
      Contingency management and deliberative decision-making processes.
      ) and have the advantage of providing a highly translational approach to this problem (
      • Bentzley B.S.
      • Fender K.M.
      • Aston-Jones G.
      The behavioral economics of drug self-administration: A review and new analytical approach for within-session procedures.
      ).
      Figure thumbnail gr1
      Figure 1Theoretical example of how a potential therapy can be developed and optimized by considering the algorithmic process that gives rise to a particular symptom or behavioral phenotype, in this case, a reward learning deficit. Slowed learning (observed behavioral phenotype) can be caused by a number of upstream differences in the reinforcement learning algorithm connecting disordered neurobiology to behavior such as reduced reward sensitivity or reduced learning rate. Selectively altering algorithmic parameters, for example, reward sensitivity in a subgroup in which reward sensitivity is the cause of the learning deficit, can change learning in a targeted way, increasing therapeutic efficacy. This type of targeted treatment assignment is not possible by considering the observed phenotype alone.
      One somewhat slippery aspect of developing descriptions at the algorithmic level is that in the brain, it is entirely possible that there are many algorithmic levels—computational procedures operating at different levels of description in the nervous system. In using algorithmic similarities to understand or treat behavioral or mental dysfunction, there are several reasons why this may be a moving target: first, there may be different levels of representation in the brain currently unknown to us; second, the processing at any level is always changing due to learning and adaptation; and third, there is also algorithmic degeneracy, in which the brain might concurrently implement different algorithms whose computational objective is similar (
      • Sajid N.
      • Parr T.
      • Hope T.M.
      • Price C.J.
      • Friston K.J.
      Degeneracy and redundancy in active inference.
      ), and which may provide the capacity for compensation in the case of functional disruption (
      • Fornito A.
      • Zalesky A.
      • Breakspear M.
      The connectomics of brain disorders.
      ). Nevertheless, the computational characteristics of psychiatric patients that are reflected in their symptoms may provide key constraints in the search for their algorithmic underpinnings. These constraints might reveal the dimensionality of the algorithmic repertoire—the number and form of the latent subprocesses that might be most important in determining pathological behavior. Overall, a broader view may be needed to determine the susceptibility of a particular algorithm to a given intervention in the context of an individual’s capacity for adaptation and/or compensation.
      The algorithmic level also remains an essential component of other types of translational research, namely, the use of model organisms. It provides a convenient currency for identifying functional similarities and differences across species, which is crucial for valid modeling of psychiatric symptoms using experimental animals (
      • Redish A.D.
      • Kepecs A.
      • Anderson L.M.
      • Calvin O.L.
      • Grissom N.M.
      • Haynos A.F.
      • et al.
      Computational validity: Using computation to translate behaviours across species.
      ). For example, the identification of similarities and differences between corticostriatal functional connectivity across species (
      • Liu X.
      • Eickhoff S.B.
      • Caspers S.
      • Wu J.
      • Genon S.
      • Hoffstaedter F.
      • et al.
      Functional parcellation of human and macaque striatum reveals human-specific connectivity in the dorsal caudate.
      ,
      • Balsters J.H.
      • Zerbi V.
      • Sallet J.
      • Wenderoth N.
      • Mars R.B.
      Primate homologs of mouse cortico-striatal circuits.
      ), when taken in isolation, can provide only limited insight into behavior. Here, analysis at the algorithmic level provides crucial complementary information. One possibility is that the same algorithm is represented in the same way using different networks across species, so different species might show a redundant representation of the same algorithm. Alternatively, homologous circuits might implement the same algorithm(s), while nonhomologous circuits would implement different algorithms. This would open the potential for degeneracy and/or genuine behavioral differences across species.

      Psychometric Considerations

      A key challenge in moving computational psychiatry research toward forward computation to clinic translation is establishing the psychometric characteristics of both parametric estimates from computational models and model-derived estimates of brain activity (
      • Hedge C.
      • Bompas A.
      • Sumner P.
      Task reliability considerations in computational psychiatry.
      ). Test-retest reliability is a primary challenge of task-related functional magnetic resonance imaging studies (
      • Hedge C.
      • Powell G.
      • Sumner P.
      The reliability paradox: Why robust cognitive tasks do not produce reliable individual differences.
      ), with some work finding very low test-retest reliability (
      • Elliott M.L.
      • Knodt A.R.
      • Ireland D.
      • Morris M.L.
      • Poulton R.
      • Ramrakha S.
      • et al.
      What is the test-retest reliability of common task-functional MRI measures? New empirical evidence and a meta-analysis.
      ,
      • Kennedy J.T.
      • Harms M.P.
      • Korucuoglu O.
      • Astafiev S.V.
      • Barch D.M.
      • Thompson W.K.
      • et al.
      Reliability and stability challenges in ABCD task fMRI data.
      ). However, reliability may be higher in some brain regions and under some conditions, e.g., within regions robustly engaged by the task; when collecting larger amounts of data (
      • Korucuoglu O.
      • Harms M.P.
      • Astafiev S.V.
      • Golosheykin S.
      • Kennedy J.T.
      • Barch D.M.
      • Anokhin A.P.
      Test-retest reliability of neural correlates of response inhibition and error monitoring: An fMRI study of a stop-signal task.
      ,
      • Korucuoglu O.
      • Harms M.P.
      • Astafiev S.V.
      • Kennedy J.T.
      • Golosheykin S.
      • Barch D.M.
      • Anokhin A.P.
      Test-retest reliability of fMRI-measured brain activity during decision making under risk.
      ,
      • Bossier H.
      • Roels S.P.
      • Seurinck R.
      • Banaschewski T.
      • Barker G.J.
      • Bokde A.L.W.
      • et al.
      The empirical replicability of task-based fMRI as a function of sample size.
      ,
      • Noble S.
      • Scheinost D.
      • Finn E.S.
      • Shen X.
      • Papademetris X.
      • McEwen S.C.
      • et al.
      Multisite reliability of MR-based functional connectivity.
      ); or when focusing on alternative metrics such as functional connectivity (
      • Noble S.
      • Scheinost D.
      • Finn E.S.
      • Shen X.
      • Papademetris X.
      • McEwen S.C.
      • et al.
      Multisite reliability of MR-based functional connectivity.
      ) and multivariate patterns (
      • Kragel P.A.
      • Han X.
      • Kraynak T.E.
      • Gianaros P.J.
      • Wager T.D.
      fMRI can be highly reliable, but it depends on what you measure.
      ).
      While there is an emerging literature on the reliability of parameter estimates derived from computational models, findings thus far have been very mixed (
      • Brown V.M.
      • Chen J.
      • Gillan C.M.
      • Price R.B.
      Improving the reliability of computational analyses: Model-based planning and its relationship with compulsivity.
      ,
      • Shahar N.
      • Hauser T.U.
      • Moutoussis M.
      • Moran R.
      • Keramati M.
      • Dolan R.J.
      NSPN consortium
      Improving the reliability of model-based decision-making estimates in the two-stage decision task with reaction-times and drift-diffusion modeling.
      ,
      • Price R.B.
      • Brown V.
      • Siegle G.J.
      Computational modeling applied to the dot-probe task yields improved reliability and mechanistic insights.
      ,
      • Weidinger L.
      • Gradassi A.
      • Molleman L.
      • van den Bos W.
      Test-retest reliability of canonical reinforcement learning models.
      ), and few systematic studies of factors that might influence test-retest reliability across different populations exist. There is also an emerging literature on the reliability of computational model–based estimates of brain activation (
      • Keren H.
      • Chen G.
      • Benson B.
      • Ernst M.
      • Leibenluft E.
      • Fox N.A.
      • et al.
      Is the encoding of Reward Prediction Error reliable during development?.
      ,
      • Wilson R.C.
      • Niv Y.
      Is model fitting necessary for model-based fMRI?.
      ). However, more work is needed in this domain, including a systematic examination of best practices for optimizing reliability and the conditions under which acceptable reliability is obtained. Further, it is important to acknowledge that psychometric characteristics are not an inherent property of a task or a model but are also dependent on the sample being examined (
      • Barch D.M.
      • Mathalon D.H.
      Using brain imaging measures in studies of procognitive pharmacologic agents in schizophrenia: Psychometric and quality assurance considerations.
      ).
      Another key challenge is making predictions about individuals rather than inferences about group-level differences in parameter estimates or brain activity. Most studies using theory-driven computational modeling approaches to examine either behavior or brain activation focus on group-level differences or correlations with symptom dimensions (
      • Hernaus D.
      • Gold J.M.
      • Waltz J.A.
      • Frank M.J.
      Impaired expected value computations coupled with overreliance on stimulus-response learning in schizophrenia.
      ,
      • Dowd E.C.
      • Frank M.J.
      • Collins A.
      • Gold J.M.
      • Barch D.M.
      Probabilistic reinforcement learning in patients with schizophrenia: Relationships to anhedonia and avolition.
      ,
      • Gold J.M.
      • Strauss G.P.
      • Waltz J.A.
      • Robinson B.M.
      • Brown J.K.
      • Frank M.J.
      Negative symptoms of schizophrenia are associated with abnormal effort-cost computations.
      ,
      • Gillan C.M.
      • Papmeyer M.
      • Morein-Zamir S.
      • Sahakian B.J.
      • Fineberg N.A.
      • Robbins T.W.
      • de Wit S.
      Disruption in the balance between goal-directed behavior and habit learning in obsessive-compulsive disorder.
      ,
      • Katthagen T.
      • Kaminski J.
      • Heinz A.
      • Buchert R.
      • Schlagenhauf F.
      Striatal dopamine and reward prediction error signaling in unmedicated schizophrenia patients.
      ,
      • Radua J.
      • Schmidt A.
      • Borgwardt S.
      • Heinz A.
      • Schlagenhauf F.
      • McGuire P.
      • Fusar-Poli P.
      Ventral striatal activation during reward processing in psychosis: A neurofunctional meta-analysis.
      ,
      • Berwian I.M.
      • Wenzel J.G.
      • Collins A.G.E.
      • Seifritz E.
      • Stephan K.E.
      • Walter H.
      • Huys Q.J.M.
      Computational mechanisms of effort and reward decisions in patients with depression and their association with relapse after antidepressant discontinuation.
      ). However, for computational psychiatry approaches to be useful for clinical application, we will need to be able to make inferences about specific individuals and to monitor longitudinal changes across time within an individual or group during treatment. Longitudinal designs raise a series of critical measurement issues, requiring not just stability of effects observable at the group level (and potentially indicative of generalizable mechanisms) but also reliability at the individual level to allow the quantification of meaningful variations in the mechanisms (
      • Hedge C.
      • Powell G.
      • Sumner P.
      The reliability paradox: Why robust cognitive tasks do not produce reliable individual differences.
      ,
      • Rouder J.N.
      • Haaf J.M.
      A psychometrics of individual differences in experimental tasks.
      ,
      • Enkavi A.Z.
      • Eisenberg I.W.
      • Bissett P.G.
      • Mazza G.L.
      • MacKinnon D.P.
      • Marsch L.A.
      • Poldrack R.A.
      Large-scale analysis of test-retest reliabilities of self-regulation measures.
      ). The extension of computational psychiatry approaches to individual level prediction and longitudinal within-person change is just in its infancy, but it is a critical pathway forward to realizing the goals of computation to clinic translation (
      • Huys Q.J.M.
      • Maia T.V.
      • Frank M.J.
      Computational psychiatry as a bridge from neuroscience to clinical applications.
      ,
      • Petzschner F.H.
      • Weber L.A.E.
      • Gard T.
      • Stephan K.E.
      Computational psychosomatics and computational psychiatry: Toward a joint framework for differential diagnosis.
      ,
      • Konova A.B.
      • Lopez-Guzman S.
      • Urmanche A.
      • Ross S.
      • Louie K.
      • Rotrosen J.
      • Glimcher P.W.
      Computational markers of risky decision-making for identification of temporal windows of vulnerability to opioid use in a real-world clinical setting.
      ,
      • Yip S.W.
      • Konova A.B.
      Densely sampled neuroimaging for maximizing clinical insight in psychiatric and addiction disorders.
      ) (see also Beyond Case Control: Capturing Dynamic Processes Via Longitudinal Computational Assessment).
      Given the lack of knowledge about basic pathophysiology underlying psychiatric disorders, determining what functions as the gold standard for determining validity is a matter of consensus among subject-matter experts. While patient self-report will in some cases be the appropriate gold standard, this may not be universally true. Medically, one example of this is the instance of referred pain: A patient may self-report left arm pain that, on examination, turns out to be caused by pain and muscle damage elsewhere, such as would be the case in a heart attack. In this instance, it is not the case that self-report is unreliable or incorrect, it is simply that it alone is not sufficient to adequately diagnose the underlying pathology. In psychiatry, DSM diagnosis is also often considered as a gold standard metric for determining validity in psychiatry (i.e., does this computational parameter have sensitivity and specificity for a given disorder). However, as has been highlighted previously, this too may be problematic, because a single diagnostic label may result from highly heterogeneous biological and computational causes.

      Transdiagnostic and Preclinical Considerations

      In disease models from medicine, a biological process needs to be measurably related to the indices of illness, and treatment needs to alter it, thereby improving symptoms of the illness. However, psychiatric illnesses are likely not of such a kind (
      • Konova A.B.
      • Lopez-Guzman S.
      • Urmanche A.
      • Ross S.
      • Louie K.
      • Rotrosen J.
      • Glimcher P.W.
      Computational markers of risky decision-making for identification of temporal windows of vulnerability to opioid use in a real-world clinical setting.
      ); with complex compensatory processes, any symptom or associated behavioral manifestation could arise from multiple and distinct underlying causes, or the same underlying cause could lead and contribute to multiple symptoms (
      • Insel T.
      • Cuthbert B.
      • Garvey M.
      • Heinssen R.
      • Pine D.S.
      • Quinn K.
      • et al.
      Research domain criteria (RDoC): Toward a new classification framework for research on mental disorders.
      ). Indeed, computational, cognitive, and learning processes have been associated with specific symptom complexes (
      • Redish A.D.
      • Kepecs A.
      • Anderson L.M.
      • Calvin O.L.
      • Grissom N.M.
      • Haynos A.F.
      • et al.
      Computational validity: Using computation to translate behaviours across species.
      ,
      • Hyman S.E.
      The diagnosis of mental disorders: The problem of reification.
      ,
      • Gillan C.M.
      • Kosinski M.
      • Whelan R.
      • Phelps E.A.
      • Daw N.D.
      Characterizing a psychiatric symptom dimension related to deficits in goal-directed control.
      ,
      • Patzelt E.H.
      • Kool W.
      • Millner A.J.
      • Gershman S.J.
      Incentives boost model-based control across a range of severity on several psychiatric constructs.
      ,
      • Amlung M.
      • Marsden E.
      • Holshausen K.
      • Morris V.
      • Patel H.
      • Vedelago L.
      • et al.
      Delay discounting as a transdiagnostic process in psychiatric disorders: A meta-analysis.
      ,
      • Gillan C.M.
      • Kalanthroff E.
      • Evans M.
      • Weingarden H.M.
      • Jacoby R.J.
      • Gershkovich M.
      • et al.
      Comparison of the association between goal-directed planning and self-reported compulsivity vs obsessive-compulsive disorder diagnosis [published correction appears in JAMA Psychiatry 2020; 77:10].
      ), and emerging evidence indicates that engaging such specific and potentially transdiagnostic markers may have clinical efficacy (
      • Drysdale A.T.
      • Grosenick L.
      • Downar J.
      • Dunlop K.
      • Mansouri F.
      • Meng Y.
      • et al.
      Resting-state connectivity biomarkers define neurophysiological subtypes of depression [published correction appears in Nat Med 2017; 23:264].
      ,
      • Pizzagalli D.A.
      • Smoski M.
      • Ang Y.S.
      • Whitton A.E.
      • Sanacora G.
      • Mathew S.J.
      • et al.
      Selective kappa-opioid antagonism ameliorates anhedonic behavior: Evidence from the Fast-fail Trial in Mood and Anxiety Spectrum Disorders (FAST-MAS) [published correction appears in Neuropsychopharmacology 2021; 46:2224].
      ,
      • Price R.B.
      • Gillan C.M.
      • Hanlon C.
      • Ferrarelli F.
      • Kim T.
      • Karim H.T.
      • et al.
      Effect of experimental manipulation of the orbitofrontal cortex on short-term markers of compulsive behavior: A theta burst stimulation study.
      ). Nevertheless, recent work has also raised questions about whether, for instance, behavioral findings can be informative about self-reported measures of illness (
      • Eisenberg I.W.
      • Bissett P.G.
      • Zeynep Enkavi A.
      • Li J.
      • MacKinnon D.P.
      • Marsch L.A.
      • Poldrack R.A.
      Uncovering the structure of self-regulation through data-driven ontology discovery.
      ).
      Transdiagnostic, dimensional approaches, such as the popular Research Domain Criteria (
      • Insel T.
      • Cuthbert B.
      • Garvey M.
      • Heinssen R.
      • Pine D.S.
      • Quinn K.
      • et al.
      Research domain criteria (RDoC): Toward a new classification framework for research on mental disorders.
      ), may be critical for integrating scientific findings across basic, preclinical, and clinical domains [for additional discussion see (
      • First M.B.
      Current state of psychiatric nosology.
      )]. However, one predominant concern given the complexity of behavior is knowing what processes we are capturing. Consider anhedonia, the inability to experience pleasure, a construct listed within the Research Domain Criteria framework as a behavioral unit of analysis reflective of negative valence systems (
      National Institute of Mental Health
      Research Domain Criteria (RDoC).
      ). In psychiatry, anhedonia is most commonly known as a core criterion for major depressive disorder, but it is also present across other psychiatric diagnoses, e.g., addiction (
      • Kiluk B.D.
      • Yip S.W.
      • DeVito E.E.
      • Carroll K.M.
      • Sofuoglu M.
      Anhedonia as a key clinical feature in the maintenance and treatment of opioid use disorder.
      ,
      • Huys Q.J.
      • Pizzagalli D.A.
      • Bogdan R.
      • Dayan P.
      Mapping anhedonia onto reinforcement learning: A behavioural meta-analysis.
      ). In humans, anhedonia involves markedly diminished interest or pleasure in all, or almost all, activities (
      American Psychiatric Association
      Diagnostic and Statistical Manual of Mental Disorders (DSM-5).
      ). This includes the inability to take interest in topics or hobbies that an individual previously found engaging and a general lack of motivation surrounding the pursuit of pleasures (e.g., food, sex). In rodents, anhedonia is often recognized as a decrease in reward-seeking actions or reward consumption and classically assessed with tests of self-stimulation or sucrose preference, respectively (
      • Slattery D.A.
      • Cryan J.F.
      Modelling depression in animals: At the interface of reward and stress pathways.
      ). We, of course, cannot equate a rat’s preference for a sucrose solution with the paralyzing anhedonia that characterizes major depressive disorder. We can, however, parse such behavior into component parts to determine whether the behavior observed is reflective of deficits in pleasure, motivation, or even learning (
      • Ward R.D.
      Methods for dissecting motivation and related psychological processes in rodents.
      ,
      • Thomsen K.R.
      Measuring anhedonia: Impaired ability to pursue, experience, and learn about reward.
      ,
      • Berridge K.C.
      Evolving concepts of emotion and motivation.
      ), all of which can be carefully parameterized with models of demand and value updating. Computational approaches, including deep phenotyping [e.g., (
      • Mathis M.W.
      • Mathis A.
      Deep learning tools for the measurement of animal behavior in neuroscience.
      )], provide an opportunity to uncover the parallel processes that go awry in animal models and contribute to psychiatric symptomatology in humans (
      • Sweis B.M.
      • Abram S.V.
      • Schmidt B.J.
      • Seeland K.D.
      • MacDonald 3rd, A.W.
      • Thomas M.J.
      • Redish A.D.
      Sensitivity to “sunk costs” in mice, rats, and humans.
      ,
      • Sweis B.M.
      • Thomas M.J.
      • Redish A.D.
      Beyond simple tests of value: Measuring addiction as a heterogeneous disease of computation-specific valuation processes.
      ,
      • Groman S.M.
      • Massi B.
      • Mathias S.R.
      • Lee D.
      • Taylor J.R.
      Model-free and model-based influences in addiction-related behaviors.
      ,
      • Groman S.M.
      • Rich K.M.
      • Smith N.J.
      • Lee D.
      • Taylor J.R.
      Chronic exposure to methamphetamine disrupts reinforcement-based decision making in rats.
      ,
      • Zhukovsky P.
      • Puaud M.
      • Jupp B.
      • Sala-Bayo J.
      • Alsiö J.
      • Xia J.
      • et al.
      Withdrawal from escalated cocaine self-administration impairs reversal learning by disrupting the effects of negative feedback on reward exploitation: A behavioral and computational analysis [published correction appears in Neuropsychopharmacology 2020; 45:714].
      ). It is nonetheless critical to note that the accuracy and utility of computational models will ultimately be dependent on the precision of the phenotyping itself (
      • Hitchcock P.F.
      • Fried E.I.
      • Frank M.J.
      Computational psychiatry needs time and context.
      ).
      Another important consideration is disease heterogeneity. For example, if individual variation is readily apparent in a given behavioral paradigm in rodents, should we expect to see such behavioral variability between human subjects on a comparable task [e.g., (
      • Sweis B.M.
      • Abram S.V.
      • Schmidt B.J.
      • Seeland K.D.
      • MacDonald 3rd, A.W.
      • Thomas M.J.
      • Redish A.D.
      Sensitivity to “sunk costs” in mice, rats, and humans.
      ,
      • Colaizzi J.M.
      • Flagel S.B.
      • Joyner M.A.
      • Gearhardt A.N.
      • Stewart J.L.
      • Paulus M.P.
      Mapping sign-tracking and goal-tracking onto human behaviors.
      ,
      • Joyner M.A.
      • Gearhardt A.N.
      • Flagel S.B.
      A translational model to assess sign-tracking and goal-tracking behavior in children.
      )]? If this variability is apparent in humans, how do we determine whether it is reflective of the same underlying processes captured in the rodent model? One recently proposed framework for considering this is that of computational validity (
      • Redish A.D.
      • Kepecs A.
      • Anderson L.M.
      • Calvin O.L.
      • Grissom N.M.
      • Haynos A.F.
      • et al.
      Computational validity: Using computation to translate behaviours across species.
      ), or the computational similarity between information processing demands underlying parallel tasks across species (
      • Redish A.D.
      • Kepecs A.
      • Anderson L.M.
      • Calvin O.L.
      • Grissom N.M.
      • Haynos A.F.
      • et al.
      Computational validity: Using computation to translate behaviours across species.
      ). Importantly, similar questions may apply to transdiagnostic human research, e.g., are the biological and cognitive substrates of anhedonia in fact shared across diagnoses? In addition, it is important to note that some computational processes (e.g., variables indexing reinforcement learning) may not consistently generalize across different contexts and thus may, in fact, be reflective of specific behaviors under study rather than a shared latent psychological construct (
      • Eckstein M.K.
      • Wilbrecht L.
      • Collins A.G.E.
      What do reinforcement learning models measure? Interpreting model parameters in cognition and neuroscience.
      ). While we cannot offer concrete answers to these emerging questions at present, we urge researchers to think deeply about such questions and, in turn, the approaches they are using both between- and within-species to advance our knowledge pertaining to translational clinical neuroscience.

      Beyond Case Control: Capturing Dynamic Processes Via Longitudinal Computational Assessment

      Clinically orientated neuroscience research has largely depended on cross-sectional designs, often comparing a group of individuals with a given condition or disorder to a group of control participants. Yet, these studies are not optimized for individual-level prediction or for the study of what are ultimately highly dynamic conditions, with varying symptom triggers between individuals. Moreover, cross-sectional markers do not necessarily hold information about longitudinal change relevant for intervention and therapy. Thus, to reach its clinical potential, future work will need to embrace the dynamic processes inherent to psychiatric symptoms and disorders by explicitly examining intraindividual longitudinal change in computational parameters and developing task paradigms and analytic methods specifically designed to capture these dynamics (
      • Yip S.W.
      • Konova A.B.
      Densely sampled neuroimaging for maximizing clinical insight in psychiatric and addiction disorders.
      ,
      • Hitchcock P.F.
      • Fried E.I.
      • Frank M.J.
      Computational psychiatry needs time and context.
      ,
      • Gueguen M.C.M.
      • Schweitzer E.M.
      • Konova A.B.
      Computational theory-driven studies of reinforcement learning and decision-making in addiction: What have we learned?.
      ). As a practical example, recent work has demonstrated that intraindividual changes in computational parameters, in this case a measure of ambiguity tolerance, precede returns to opioid use among individuals in treatment (
      • Konova A.B.
      • Lopez-Guzman S.
      • Urmanche A.
      • Ross S.
      • Louie K.
      • Rotrosen J.
      • Glimcher P.W.
      Computational markers of risky decision-making for identification of temporal windows of vulnerability to opioid use in a real-world clinical setting.
      ).
      Naturalistic changes in clinical symptoms can suggest potential novel treatment targets (
      • van Borkulo C.
      • Boschloo L.
      • Borsboom D.
      • Penninx B.W.J.H.
      • Waldorp L.J.
      • Schoevers R.A.
      Association of symptom network structure with the course of depression [published correction appears in JAMA Psychiatry 2016; 73:412].
      ), and studies examining treatments or interventions have the potential to shape clinical decision making. For instance, recent successes include the application of computational approaches to decision making (
      • Berwian I.M.
      • Wenzel J.G.
      • Collins A.G.E.
      • Seifritz E.
      • Stephan K.E.
      • Walter H.
      • Huys Q.J.M.
      Computational mechanisms of effort and reward decisions in patients with depression and their association with relapse after antidepressant discontinuation.
      ,
      • Konova A.B.
      • Lopez-Guzman S.
      • Urmanche A.
      • Ross S.
      • Louie K.
      • Rotrosen J.
      • Glimcher P.W.
      Computational markers of risky decision-making for identification of temporal windows of vulnerability to opioid use in a real-world clinical setting.
      ,
      • Chen H.
      • Nebe S.
      • Mojtahedzadeh N.
      • Kuitunen-Paul S.
      • Garbusow M.
      • Schad D.J.
      • et al.
      Susceptibility to interference between Pavlovian and instrumental control is associated with early hazardous alcohol use.
      ) and imaging (
      • Wu W.
      • Zhang Y.
      • Jiang J.
      • Lucas M.V.
      • Fonzo G.A.
      • Rolle C.E.
      • et al.
      An electroencephalographic signature predicts antidepressant response in major depression.
      ,
      • Yip S.W.
      • Scheinost D.
      • Potenza M.N.
      • Carroll K.M.
      Connectome-based prediction of cocaine abstinence.
      ,
      • Lichenstein S.D.
      • Scheinost D.
      • Potenza M.N.
      • Carroll K.M.
      • Yip S.W.
      Dissociable neural substrates of opioid and cocaine use identified via connectome-based modelling.
      ) in predicting treatment response and course in depression and substance use disorders. Although a single time-point measurement may be sufficient for diagnostic classification, and in some cases for treatment selection, other clinically relevant outcomes such as longer-term prognosis and determining if a current treatment is sufficiently working for an individual, will require denser sampling of behavior and neural function. In fact, it can be argued that most clinical decisions require continuously (re)assessing the person in time. This is most evident in managing rapidly changing clinical phenomena, such as suicidal behavior, manic/depressive episodes, and relapse to drug use, all of which require ongoing treatment modification. Even diagnosis and treatment selection can be improved upon by longitudinal data (
      • Budde M.
      • Anderson-Schmidt H.
      • Gade K.
      • Reich-Erkelenz D.
      • Adorjan K.
      • Kalman J.L.
      • et al.
      A longitudinal approach to biological psychiatric research: The PsyCourse study.
      ). A move toward person-centered computational psychiatry research is needed not only for enhancing the potential for clinical translation but also for basic research. To elucidate the mechanisms of disease, computational psychiatry efforts need to be geared more accurately toward evaluating which are the most clinically relevant (and thus most defining) algorithmic parameters and at which timescale.
      The shift toward dynamic assessment can also facilitate building computational cognitive neuroscience into the development of just-in-time adaptive interventions, which are increasing in use (
      • Coppersmith D.D.L.
      • Dempsey W.
      • Kleiman E.
      • Bentley K.
      • Murphy S.
      • Nock M.
      Just-in-time adaptive interventions for suicide prevention: Promise, challenges, and future directions.
      ,
      • Nahum-Shani I.
      • Smith S.N.
      • Spring B.J.
      • Collins L.M.
      • Witkiewitz K.
      • Tewari A.
      • Murphy S.A.
      Just-in-time adaptive interventions (JITAIs) in mobile health: Key components and design principles for ongoing health behavior support.
      ,
      • Carpenter S.M.
      • Menictas M.
      • Nahum-Shani I.
      • Wetter D.W.
      • Murphy S.A.
      Developments in mobile health just-in-time adaptive interventions for addiction science.
      ,
      • Hébert E.T.
      • Ra C.K.
      • Alexander A.C.
      • Helt A.
      • Moisiuc R.
      • Kendzor D.E.
      • et al.
      A mobile just-in-time adaptive intervention for smoking cessation: Pilot randomized controlled trial.
      ,
      • Nahum-Shani I.
      • Hekler E.B.
      • Spruijt-Metz D.
      Building health behavior models to guide the development of just-in-time adaptive interventions: A pragmatic framework.
      ). The idea to study patient behavior longitudinally has been embraced in recent years with the increased availability of cost-effective and remote data collection tools (e.g., http://www.thegreatbrainexperiment.com; https://brainexplorer.net/; https://www.neureka.ie/) or, more directly, using ecologic momentary assessments (
      • Kleiman E.M.
      • Nock M.K.
      Real-time assessment of suicidal thoughts and behaviors.
      ,
      • Ebner-Priemer U.W.
      • Trull T.J.
      Ecological momentary assessment of mood disorders and mood dysregulation.
      ,
      • de Girolamo G.
      • Barattieri di San Pietro C.
      • Bulgari V.
      • Dagani J.
      • Ferrari C.
      • Hotopf M.
      • et al.
      The acceptability of real-time health monitoring among community participants with depression: A systematic review and meta-analysis of the literature.
      ,
      • Businelle M.S.
      • Hébert E.T.
      • Kendzor D.E.
      Introduction to the special issue on use of mobile technology for real-time assessment and treatment of substance-use disorders.
      ), but only recently has increased emphasis been placed on acquiring similarly densely sampled neural measurements (
      • Laumann T.O.
      • Gordon E.M.
      • Adeyemo B.
      • Snyder A.Z.
      • Joo S.J.
      • Chen M.Y.
      • et al.
      Functional system and areal organization of a highly sampled individual human brain.
      ,
      • Poldrack R.A.
      • Laumann T.O.
      • Koyejo O.
      • Gregory B.
      • Hover A.
      • Chen M.Y.
      • et al.
      Long-term neural and physiological phenotyping of a single human.
      ,
      • Naselaris T.
      • Allen E.
      • Kay K.
      Extensive sampling for complete models of individual brains.
      ,
      • Yang Z.
      • Lewis L.D.
      Imaging the temporal dynamics of brain states with highly sampled fMRI.
      ,
      • Newbold D.J.
      • Dosenbach N.U.F.
      Tracking plasticity of individual human brains.
      ), and this work has overwhelmingly focused on the monitoring of healthy states. The neural dynamics of changing clinical phenomena remain largely unknown (
      • Yip S.W.
      • Konova A.B.
      Densely sampled neuroimaging for maximizing clinical insight in psychiatric and addiction disorders.
      ). Using computational approaches to understand these neural dynamics can help bridge between changes in computations inferred from behavior and variation in symptom expression.
      Given the heterogeneity and complexity of clinical phenomena, these longitudinal behavioral and neural dynamics should aim to capture multidimensional computational mechanisms (
      • Moutoussis M.
      • Garzón B.
      • Neufeld S.
      • Bach D.R.
      • Rigoli F.
      • Goodyer I.
      • et al.
      Decision-making ability, psychopathology, and brain connectivity.
      ). Prior work shows that conceptually distinct computational markers, such as those that describe people’s preferences for risk and delay and their propensity to learn in a model-free versus model-based way, define distinct diagnostic categories (
      • Ahn W.Y.
      • Vasilev G.
      • Lee S.H.
      • Busemeyer J.R.
      • Kruschke J.K.
      • Bechara A.
      • Vassileva J.
      Decision-making in stimulant and opiate addicts in protracted abstinence: Evidence from computational modeling with pure users.
      ,
      • Charpentier C.J.
      • Aylward J.
      • Roiser J.P.
      • Robinson O.J.
      Enhanced risk aversion, but not loss aversion, in unmedicated pathological anxiety.
      ), symptom domains (
      • Gillan C.M.
      • Kosinski M.
      • Whelan R.
      • Phelps E.A.
      • Daw N.D.
      Characterizing a psychiatric symptom dimension related to deficits in goal-directed control.
      ,
      • Wise T.
      • Dolan R.J.
      Associations between aversive learning processes and transdiagnostic psychiatric symptoms in a general population sample.
      ,
      • Rouault M.
      • Seow T.
      • Gillan C.M.
      • Fleming S.M.
      Psychiatric symptom dimensions are associated with dissociable shifts in metacognition but not task performance.
      ,
      • Baker S.C.
      • Konova A.B.
      • Daw N.D.
      • Horga G.
      A distinct inferential mechanism for delusions in schizophrenia.
      ), and clinically relevant outcomes (
      • Konova A.B.
      • Lopez-Guzman S.
      • Urmanche A.
      • Ross S.
      • Louie K.
      • Rotrosen J.
      • Glimcher P.W.
      Computational markers of risky decision-making for identification of temporal windows of vulnerability to opioid use in a real-world clinical setting.
      ,
      • Yip S.W.
      • Scheinost D.
      • Potenza M.N.
      • Carroll K.M.
      Connectome-based prediction of cocaine abstinence.
      ,
      • Lichenstein S.D.
      • Scheinost D.
      • Potenza M.N.
      • Carroll K.M.
      • Yip S.W.
      Dissociable neural substrates of opioid and cocaine use identified via connectome-based modelling.
      ). Further, even conceptually related markers, such as preferences for known risk and unknown or ambiguous risk, can be differentially predictive of the same clinical outcome (
      • Konova A.B.
      • Lopez-Guzman S.
      • Urmanche A.
      • Ross S.
      • Louie K.
      • Rotrosen J.
      • Glimcher P.W.
      Computational markers of risky decision-making for identification of temporal windows of vulnerability to opioid use in a real-world clinical setting.
      ). Thus, focusing on just a few behavioral and neural variables at a time could preclude more detailed computational consideration of related processes and heterogeneous clinical profiles. A longitudinal and multidimensional examination, a type of dynamic neurocomputational fingerprinting approach (
      • Gueguen M.C.M.
      • Schweitzer E.M.
      • Konova A.B.
      Computational theory-driven studies of reinforcement learning and decision-making in addiction: What have we learned?.
      ), may therefore provide a more complete understanding of mental illness and aid in developing better tailored and timed interventions at an individual level or, perhaps more immediately, at the group level via patient stratification.

      The Final Frontier: Clinical Challenges to Implementation

      The final hurdles for effective translation from computation to clinic are of course those directly pertaining to implementation and subsequent treatment development and selection: how can we develop a pragmatic framework that would enable the development of computationally informed tests in psychiatry for different mental health conditions?
      The types of outcomes under consideration for a test are highly variable based on time frame, level of observation (e.g., symptoms vs. biological process), and interventions available. A screening test would be useful to identify individuals at risk for the disorder or who have a not-yet-clinically manifest disorder (
      • Maxim L.D.
      • Niebo R.
      • Utell M.J.
      Screening tests: A review with examples [published correction appears in Inhal Toxicol 2019; 31:298].
      ,
      • Wilson J.M.G.
      • Jungner G.
      Principles and Practice of Screening for Disease.
      ). A diagnostic test would provide evidence for the presence of a particular disease or help arbitrate between diseases with similar manifestations (
      • Ferrante di Ruffano L.
      • Hyde C.J.
      • McCaffery K.J.
      • Bossuyt P.M.M.
      • Deeks J.J.
      Assessing the value of diagnostic tests: A framework for designing and evaluating trials.
      ). A prognostic test would provide patients and providers with information regarding degree of recovery, severity of residual symptoms, occurrence of associated complications, or likelihood of a disease-free interval (
      ,
      • Rector T.S.
      • Taylor B.C.
      • Wilt T.J.
      Chapter 12: Systematic review of prognostic tests.
      ). Finally, a treatment-specific test would help to guide which type of intervention is most likely going to be associated with a positive outcome for a particular individual. Within a computational framework, this means that one will need to explicitly consider how risk, diagnoses, recurrence, or recovery translate to model parameters that can give insights to the mechanistic aspect of the disorder as well as pragmatically be as robust and reliable as clinical tests.
      Another pragmatic consideration is to determine who benefits from the test. A test would provide a patient with information that can be used to adjust activities, treatment selection, and adherence and integrate individual experiences into an explanatory disease model. The provider benefits from a test by having a more precise disease model, selecting disease-specific interventions, based on the underlying algorithmic disruption, and focusing attention on monitoring disease-specific outcomes. From a payer perspective, tests—even if not sufficiently sensitive or specific for individual cases—can aid in deployment of resources to a particular intervention or disease or other operational decisions. Finally, a public health specialist can use tests to determine the need for population-based resource allocation to reduce disease impact.
      Test characteristics will have different implications for each of these stakeholders. An important challenge is to translate computational models to stakeholders such that they become both understood and actionable. Specifically, using a reinforcement model framework, the notion of different learning rates for gains or losses as critical parameters for mood disorders may require experts to reframe these measures in terms of past history considered when taking rewards or losses into account. However, an actionable test alone may also prove useful (e.g., an aggregate risk calculator approach, which is used to trigger adjustment of a well-validated treatment). Moreover, the net benefit of a computationally informed test needs to be expressed in numbers that are meaningful to stakeholders (Figures 2 and 3). For example, the number of recurrences of a depressive episode that could be prevented with a positive test may be useful for a public health specialist, whereas the likelihood of a particular patient experiencing a depressive episode within the next 6 months may be more useful for a clinician.
      Figure thumbnail gr2
      Figure 2Incremental cost-effectiveness ratio (ICER) analysis. ICER analysis is a method of associating an economic value to the cost of conducting the test that can quantitatively estimate the effect that a test-based assignment can have on treatment. ICER is based on the incremental costs per unit of effect of the intervention and calculated as the difference in the sum of specific direct and indirect costs divided by the inverted difference in effect score [e.g., (
      • Müller G.
      • Pfinder M.
      • Schmahl C.
      • Bohus M.
      • Lyssenko L.
      Cost-effectiveness of a mindfulness-based mental health promotion program: Economic evaluation of a nonrandomized controlled trial with propensity score matching.
      )]. In the example shown, the probability of a cost-effective clinically meaningful response to different clinical actions are plotted against the stakeholders’ willingness to pay. Willingness to pay increases with buprenorphine given the high evidence base for its utility in treating opioid use disorder. In contrast, willingness to pay decreases for nonevidence based (e.g., brief intervention) and nontreatment (i.e., referral to alternate source) actions.
      Figure thumbnail gr3
      Figure 3Schematic diagram of computational testing steps for determining clinical treatment. Conceptual overview of computational testing steps for determining clinical treatment within the specific context of tests focused on screening, treatment selection, and prognosis. However, these approaches can be readily extended to include diagnostic tests if there are differential actionables associated with a particular diagnosis. The size of the boxes and the subgroups are meant to indicate the scale of the base rate or pre- and posttest probability. DALY, disability-adjusted life years; Non-Resp., nonresponder; TX, treatment.
      A test cannot be separated from the setting in which it is administered. Even with a point prevalence of 1 in 5 individuals having a mental health condition (
      Center for Behavioral Health Statistics and Quality
      ), a test with a positive likelihood ratio of 4 to 1 (i.e., the chance that a person with a disease will have a positive test over a person that does not have the disease but will test positive) and a negative likelihood ratio of 2 to 5 (i.e., the chance that a person who does have a disease will test negative over the chance that a person who does not have the disease will test negative) would result in as many false positive as true positive cases. In other words, as many individuals would be identified as having the condition who in fact do not have it as would be identified as truly having the disease. Thus, test characteristics, i.e., specificity and sensitivity, are very much population-specific and may not hold for the population at hand. For example, assuming that the overall goal of a low-cost intervention is (at a minimum) harm reduction, a test with high sensitivity and low specificity might be acceptable for determining the need for an intervention, such as initiation of methadone for opioid use, but might not be optimal for predicting termination of methadone treatment. In both cases, the goal is relapse and overdose prevention; however, maximizing sensitivity for methadone initiation is most likely to minimize overdose risk, whereas maximizing specificity for methadone cessation is most likely to minimize overdose risk (
      • Yip S.W.
      • Kiluk B.
      • Scheinost D.
      Toward addiction prediction: An overview of cross-validated predictive modeling findings and considerations for future neuroimaging research.
      ). Within this context, a test with low sensitivity but high specificity might be acceptable for prediction of treatment cessation but not for prediction of treatment initiation (
      • Yip S.W.
      • Kiluk B.
      • Scheinost D.
      Toward addiction prediction: An overview of cross-validated predictive modeling findings and considerations for future neuroimaging research.
      ). It is equally important to note that a test evaluated in the general population may behave very differently in a clinical population of a provider. Thus, the population in which a computational model is developed and tested is also essential to consider.
      Tests are frequently evaluated by their statistical characteristics, which some have called the single most problematic misrepresentation of the utility of a test (
      ). Sensitivity and specificity do not provide sufficient information to judge the utility of a test because they do not consider the base rate of the disease (or the pretest probability, in Bayesian terms). Positive and negative likelihood ratios can readily be used to compute posttest probabilities, which provide an intuitive notion of the certainty a test can provide. Nevertheless, even these numbers are insufficient to readily assess the test utility for 2 reasons. First, there are pragmatic aspects of a test that are not reflected in these numbers. As mentioned above, the base rate of the disease in the population tested is one aspect, but more importantly, does the test involve an individual assessment by a trained provider, is it dependent on its implementation, and/or does the test itself alter the disease state in the individual or change behavior subsequently? These issues are particularly relevant for mental health conditions. Second, what are the interventions associated with a positive or negative test, and how do costs, intensity of intervention, and probability of successful or unsuccessful outcome change because of the test result? In summary, a test is merely a step embedded in a chain of evaluations and interventions aimed to improve patient outcome and needs to be evaluated as such.
      These ideas might be integrated with potential for in silico simulation for the development and optimization of treatments at the algorithmic level (see From the Bottom Up: Toward Algorithmic Development of Novel Therapies). Specifically, algorithmic models of psychiatric dysfunction could be used to simulate clinical status and predicted clinical course if the mathematical form of these can be estimated. Presumably, at first, initial tests of symptoms (e.g., self-report diagnosis) may provide a somewhat uncertain indication of a patient’s clinical course, but further evaluation and tests may provide more accuracy and, in particular, favor one model over another (i.e., provide more accurate differential diagnosis). Monitoring the model predictions could be conducted with further testing during treatment, and predictions of the model could be refined further. Typically, it would be expected that a single model be selected with high likelihood, and this would describe the clinical course with high accuracy. In practice, however, there might be multiple plausible models that will describe a participant’s symptoms, and these might be difficult to differentiate with available tests. Treatments that might cause harm under one plausible scenario might be avoided, while the probability of different scenarios might be used for weighting the utility of tests or treatments in light of information about costs and benefits. Further testing should eventually disambiguate the different models according to the precision of their predictions, and oversight by a physician will be particularly important in cases where model error is particularly high (for example, a manic episode elicited by an antidepressant medication during treatment). Overall, this system might be implemented computationally as a mixture of experts approach (
      • Yuksel S.E.
      • Wilson J.N.
      • Gader P.D.
      Twenty years of mixture of experts.
      ,
      • Jacobs R.A.
      • Jordan M.I.
      • Nowlan S.J.
      • Hinton G.E.
      Adaptive mixtures of local experts.
      ), in which different models can be gracefully integrated to provide unique predictions across different domains (e.g., across uniquely specified disorders) or compete to describe a given domain. This framework could provide weighting for probabilities of different scenarios, as well as utilities for potential harms, costs, and clinical benefits. The physician is also represented within this system, both as a provider of information (e.g., through the diagnosis) and also as an independent expert with their own biases (
      • Downar J.
      • Bhatt M.
      • Montague P.R.
      Neural correlates of effective learning in experienced medical decision-makers.
      ), but who might exert more control if the model predictions are associated with risk or are inaccurate. The computational researchers who derived the test are also inherently represented within this system; thus, for maximal translation from computation to clinic (as opposed to vice versa), even basic computational research must also weigh the above considerations.

      Conclusions

      Effective forward translation from computation to clinic remains elusive, yet it may be enabled by careful consideration of mechanisms at an algorithmic level, psychometric standardization, recent developments in longitudinal phenotyping and other theory-driven computational approaches, and careful, realistic evaluation of a test’s efficacy within a specific real-world clinical context.

      Acknowledgments and Disclosures

      This work has been supported in part by the William K. Warren Foundation, the National Institute on Drug Abuse (Grant No. U01 DA041089 ), and the National Institute of General Medical Sciences Center (Grant No. 1P20GM121312 [to MP]). SF was funded, in part, by the Pritzker Neuropsychiatric Disorders Research Consortium Fund LLC (http://www.pritzkerneuropsych.org). ABK was funded by the National Institute on Drug Abuse (Grant Nos. R01DA053282 and R01DA054201 ). DMB was funded by the National Institutes of Health (Grant Nos. R01 MH066031 and R01 MH084840 ). SWY was funded in part by the National Institute on Alcohol Abuse and Alcoholism (Grant No. R01 AA027553 ).
      QJMH acknowledges support from the University College London Hospital National Institute for Health and Care Biomedical Research Centre and has received consultancy fees and options from Aya Health and Alto Neurosciences and a research grant from Koa Health. DMB receives grant funding from the National Institute of Mental Health and the National Institute on Drug Abuse and consults for Boehringer Ingelheim on mobile sensing approaches to assessment. MP is an adviser to Spring Care, Inc., a behavioral health startup and has received royalties for an article about methamphetamine in UpToDate. All other authors report no biomedical financial interest or potential conflicts of interest.

      References

        • Yip S.W.
        • Kiluk B.
        • Scheinost D.
        Toward addiction prediction: An overview of cross-validated predictive modeling findings and considerations for future neuroimaging research.
        Biol Psychiatry Cogn Neurosci Neuroimaging. 2020; 5: 748-758
        • Cohrs R.J.
        • Martin T.
        • Ghahramani P.
        • Bidaut L.
        • Higgins P.J.
        • Shahzad A.
        Translational medicine definition by the European Society for Translational Medicine.
        New Horiz Transl Med. 2015; 2: 86-88
        • Huys Q.J.M.
        • Maia T.V.
        • Frank M.J.
        Computational psychiatry as a bridge from neuroscience to clinical applications.
        Nat Neurosci. 2016; 19: 404-413
        • Marr D.
        • Poggio T.
        From understanding computation to understanding neural circuitry: Artificial Intelligence Laboratory AI Memo 357.
        Massachusetts Institute of Technology, Cambridge, Massachusetts1976
        • Haber S.N.
        The place of dopamine in the cortico-basal ganglia circuit.
        Neuroscience. 2014; 282: 248-257
        • Schultz W.
        • Dayan P.
        • Montague P.R.
        A neural substrate of prediction and reward.
        Science. 1997; 275: 1593-1599
        • Nair A.
        • Rutledge R.B.
        • Mason L.
        Under the hood: Using computational psychiatry to make psychological therapies more mechanism-focused.
        Front Psychiatry. 2020; 11: 140
        • Robinson E.
        Psychopharmacology: From serendipitous discoveries to rationale design, but what next?.
        Brain Neurosci Adv. 2018; 2 (2398212818812629)
        • Kapur S.
        • Mamo D.
        Half a century of antipsychotics and still a central role for dopamine D2 receptors.
        Prog Neuropsychopharmacol Biol Psychiatry. 2003; 27: 1081-1090
        • Braver T.S.
        • Barch D.M.
        • Cohen J.D.
        Cognition and control in schizophrenia: A computational model of dopamine and prefrontal function.
        Biol Psychiatry. 1999; 46: 312-328
        • Maia T.V.
        • Frank M.J.
        An integrative perspective on the role of dopamine in schizophrenia.
        Biol Psychiatry. 2017; 81: 52-66
        • Smith A.J.
        • Li M.
        • Becker S.
        • Kapur S.
        Linking animal models of psychosis to computational models of dopamine function.
        Neuropsychopharmacology. 2007; 32: 54-66
        • Heinz A.
        • Murray G.K.
        • Schlagenhauf F.
        • Sterzer P.
        • Grace A.A.
        • Waltz J.A.
        Towards a unifying cognitive, neurophysiological, and computational neuroscience account of schizophrenia.
        Schizophr Bull. 2019; 45: 1092-1100
        • Adams R.A.
        • Stephan K.E.
        • Brown H.R.
        • Frith C.D.
        • Friston K.J.
        The computational anatomy of psychosis.
        Front Psychiatry. 2013; 4: 47
        • Bickel W.K.
        • Johnson M.W.
        • Koffarnus M.N.
        • MacKillop J.
        • Murphy J.G.
        The behavioral economics of substance use disorders: Reinforcement pathologies and their repair.
        Annu Rev Clin Psychol. 2014; 10: 641-677
        • Peters J.
        • Miedl S.F.
        • Büchel C.
        Formal comparison of dual-parameter temporal discounting models in controls and pathological gamblers.
        PLoS One. 2012; 7e47225
        • Stein J.S.
        • Wilson A.G.
        • Koffarnus M.N.
        • Daniel T.O.
        • Epstein L.H.
        • Bickel W.K.
        Unstuck in time: Episodic future thinking reduces delay discounting and cigarette smoking.
        Psychopharmacology (Berl). 2016; 233: 3771-3778
        • Bickel W.K.
        • Yi R.
        • Landes R.D.
        • Hill P.F.
        • Baxter C.
        Remember the future: Working memory training decreases delay discounting among stimulant addicts.
        Biol Psychiatry. 2011; 69: 260-265
        • González-Roz A.
        • Jackson J.
        • Murphy C.
        • Rohsenow D.J.
        • MacKillop J.
        Behavioral economic tobacco demand in relation to cigarette consumption and nicotine dependence: A meta-analysis of cross-sectional relationships.
        Addiction. 2019; 114: 1926-1940
        • Hursh S.R.
        • Silberberg A.
        Economic demand and essential value.
        Psychol Rev. 2008; 115: 186-198
        • MacKillop J.
        • Few L.R.
        • Murphy J.G.
        • Wier L.M.
        • Acker J.
        • Murphy C.
        • et al.
        High-resolution behavioral economic analysis of cigarette demand to inform tax policy.
        Addiction. 2012; 107: 2191-2200
        • González-Roz A.
        • García-Pérez Á.
        • Weidberg S.
        • Aonso-Diego G.
        • Secades-Villa R.
        Reinforcer pathology and response to contingency management for smoking cessation.
        Psychol Addict Behav. 2020; 34: 156-163
        • Yoon J.H.
        • Suchting R.
        • McKay S.A.
        • San Miguel G.G.
        • Vujanovic A.A.
        • Stotts A.L.
        • et al.
        Baseline cocaine demand predicts contingency management treatment outcomes for cocaine-use disorder.
        Psychol Addict Behav. 2020; 34: 164-174
        • Yoon J.H.
        • Suchting R.
        • de Dios C.
        • Vincent J.N.
        • McKay S.A.
        • Lane S.D.
        • Schmitz J.M.
        Decreased cocaine demand following contingency management treatment.
        Drug Alcohol Depend. 2021; 226: 108883
        • Regier P.S.
        • Redish A.D.
        Contingency management and deliberative decision-making processes.
        Front Psychiatry. 2015; 6: 76
        • Bentzley B.S.
        • Fender K.M.
        • Aston-Jones G.
        The behavioral economics of drug self-administration: A review and new analytical approach for within-session procedures.
        Psychopharmacology (Berl). 2013; 226: 113-125
        • Sajid N.
        • Parr T.
        • Hope T.M.
        • Price C.J.
        • Friston K.J.
        Degeneracy and redundancy in active inference.
        Cereb Cortex. 2020; 30: 5750-5766
        • Fornito A.
        • Zalesky A.
        • Breakspear M.
        The connectomics of brain disorders.
        Nat Rev Neurosci. 2015; 16: 159-172
        • Redish A.D.
        • Kepecs A.
        • Anderson L.M.
        • Calvin O.L.
        • Grissom N.M.
        • Haynos A.F.
        • et al.
        Computational validity: Using computation to translate behaviours across species.
        Philos Trans R Soc Lond B Biol Sci. 2022; 377: 20200525
        • Liu X.
        • Eickhoff S.B.
        • Caspers S.
        • Wu J.
        • Genon S.
        • Hoffstaedter F.
        • et al.
        Functional parcellation of human and macaque striatum reveals human-specific connectivity in the dorsal caudate.
        Neuroimage. 2021; 235: 118006
        • Balsters J.H.
        • Zerbi V.
        • Sallet J.
        • Wenderoth N.
        • Mars R.B.
        Primate homologs of mouse cortico-striatal circuits.
        Elife. 2020; 9e536820
        • Hedge C.
        • Bompas A.
        • Sumner P.
        Task reliability considerations in computational psychiatry.
        Biol Psychiatry Cogn Neurosci Neuroimaging. 2020; 5: 837-839
        • Hedge C.
        • Powell G.
        • Sumner P.
        The reliability paradox: Why robust cognitive tasks do not produce reliable individual differences.
        Behav Res Methods. 2018; 50: 1166-1186
        • Elliott M.L.
        • Knodt A.R.
        • Ireland D.
        • Morris M.L.
        • Poulton R.
        • Ramrakha S.
        • et al.
        What is the test-retest reliability of common task-functional MRI measures? New empirical evidence and a meta-analysis.
        Psychol Sci. 2020; 31: 792-806
        • Kennedy J.T.
        • Harms M.P.
        • Korucuoglu O.
        • Astafiev S.V.
        • Barch D.M.
        • Thompson W.K.
        • et al.
        Reliability and stability challenges in ABCD task fMRI data.
        Neuroimage. 2022; 252: 199046
        • Korucuoglu O.
        • Harms M.P.
        • Astafiev S.V.
        • Golosheykin S.
        • Kennedy J.T.
        • Barch D.M.
        • Anokhin A.P.
        Test-retest reliability of neural correlates of response inhibition and error monitoring: An fMRI study of a stop-signal task.
        Front Neurosci. 2021; 15: 624911
        • Korucuoglu O.
        • Harms M.P.
        • Astafiev S.V.
        • Kennedy J.T.
        • Golosheykin S.
        • Barch D.M.
        • Anokhin A.P.
        Test-retest reliability of fMRI-measured brain activity during decision making under risk.
        Neuroimage. 2020; 214: 116759
        • Bossier H.
        • Roels S.P.
        • Seurinck R.
        • Banaschewski T.
        • Barker G.J.
        • Bokde A.L.W.
        • et al.
        The empirical replicability of task-based fMRI as a function of sample size.
        Neuroimage. 2020; 212: 116601
        • Noble S.
        • Scheinost D.
        • Finn E.S.
        • Shen X.
        • Papademetris X.
        • McEwen S.C.
        • et al.
        Multisite reliability of MR-based functional connectivity.
        Neuroimage. 2017; 146: 959-970
        • Kragel P.A.
        • Han X.
        • Kraynak T.E.
        • Gianaros P.J.
        • Wager T.D.
        fMRI can be highly reliable, but it depends on what you measure.
        PsyArXiv. 2020; https://doi.org/10.31234/osf.io/9eaxk
        • Brown V.M.
        • Chen J.
        • Gillan C.M.
        • Price R.B.
        Improving the reliability of computational analyses: Model-based planning and its relationship with compulsivity.
        Biol Psychiatry Cogn Neurosci Neuroimaging. 2020; 5: 601-609
        • Shahar N.
        • Hauser T.U.
        • Moutoussis M.
        • Moran R.
        • Keramati M.
        • Dolan R.J.
        • NSPN consortium
        Improving the reliability of model-based decision-making estimates in the two-stage decision task with reaction-times and drift-diffusion modeling.
        PLOS Comput Biol. 2019; 15e1006803
        • Price R.B.
        • Brown V.
        • Siegle G.J.
        Computational modeling applied to the dot-probe task yields improved reliability and mechanistic insights.
        Biol Psychiatry. 2019; 85: 606-612
        • Weidinger L.
        • Gradassi A.
        • Molleman L.
        • van den Bos W.
        Test-retest reliability of canonical reinforcement learning models.
        in: Presented at the Conference on Cognitive Computational Neurosciences, Sep 13–16, Berlin, Germany. 2019
        • Keren H.
        • Chen G.
        • Benson B.
        • Ernst M.
        • Leibenluft E.
        • Fox N.A.
        • et al.
        Is the encoding of Reward Prediction Error reliable during development?.
        Neuroimage. 2018; 178: 266-276
        • Wilson R.C.
        • Niv Y.
        Is model fitting necessary for model-based fMRI?.
        PLoS Comput Biol. 2015; 11e1004237
        • Barch D.M.
        • Mathalon D.H.
        Using brain imaging measures in studies of procognitive pharmacologic agents in schizophrenia: Psychometric and quality assurance considerations.
        Biol Psychiatry. 2011; 70: 13-18
        • Hernaus D.
        • Gold J.M.
        • Waltz J.A.
        • Frank M.J.
        Impaired expected value computations coupled with overreliance on stimulus-response learning in schizophrenia.
        Biol Psychiatry Cogn Neurosci Neuroimaging. 2018; 3: 916-926
        • Dowd E.C.
        • Frank M.J.
        • Collins A.
        • Gold J.M.
        • Barch D.M.
        Probabilistic reinforcement learning in patients with schizophrenia: Relationships to anhedonia and avolition.
        Biol Psychiatry Cogn Neurosci Neuroimaging. 2016; 1: 460-473
        • Gold J.M.
        • Strauss G.P.
        • Waltz J.A.
        • Robinson B.M.
        • Brown J.K.
        • Frank M.J.
        Negative symptoms of schizophrenia are associated with abnormal effort-cost computations.
        Biol Psychiatry. 2013; 74: 130-136
        • Gillan C.M.
        • Papmeyer M.
        • Morein-Zamir S.
        • Sahakian B.J.
        • Fineberg N.A.
        • Robbins T.W.
        • de Wit S.
        Disruption in the balance between goal-directed behavior and habit learning in obsessive-compulsive disorder.
        Am J Psychiatry. 2011; 168: 718-726
        • Katthagen T.
        • Kaminski J.
        • Heinz A.
        • Buchert R.
        • Schlagenhauf F.
        Striatal dopamine and reward prediction error signaling in unmedicated schizophrenia patients.
        Schizophr Bull. 2020; 46: 1535-1546
        • Radua J.
        • Schmidt A.
        • Borgwardt S.
        • Heinz A.
        • Schlagenhauf F.
        • McGuire P.
        • Fusar-Poli P.
        Ventral striatal activation during reward processing in psychosis: A neurofunctional meta-analysis.
        JAMA Psychiatry. 2015; 72: 1243-1251
        • Berwian I.M.
        • Wenzel J.G.
        • Collins A.G.E.
        • Seifritz E.
        • Stephan K.E.
        • Walter H.
        • Huys Q.J.M.
        Computational mechanisms of effort and reward decisions in patients with depression and their association with relapse after antidepressant discontinuation.
        JAMA Psychiatry. 2020; 77: 513-522
        • Rouder J.N.
        • Haaf J.M.
        A psychometrics of individual differences in experimental tasks.
        Psychon Bull Rev. 2019; 26: 452-467
        • Enkavi A.Z.
        • Eisenberg I.W.
        • Bissett P.G.
        • Mazza G.L.
        • MacKinnon D.P.
        • Marsch L.A.
        • Poldrack R.A.
        Large-scale analysis of test-retest reliabilities of self-regulation measures.
        Proc Natl Acad Sci U S A. 2019; 116: 5472-5477
        • Petzschner F.H.
        • Weber L.A.E.
        • Gard T.
        • Stephan K.E.
        Computational psychosomatics and computational psychiatry: Toward a joint framework for differential diagnosis.
        Biol Psychiatry. 2017; 82: 421-430
        • Konova A.B.
        • Lopez-Guzman S.
        • Urmanche A.
        • Ross S.
        • Louie K.
        • Rotrosen J.
        • Glimcher P.W.
        Computational markers of risky decision-making for identification of temporal windows of vulnerability to opioid use in a real-world clinical setting.
        JAMA Psychiatry. 2020; 77: 368-377
        • Yip S.W.
        • Konova A.B.
        Densely sampled neuroimaging for maximizing clinical insight in psychiatric and addiction disorders.
        Neuropsychopharmacology. 2022; 47: 395-396
        • Insel T.
        • Cuthbert B.
        • Garvey M.
        • Heinssen R.
        • Pine D.S.
        • Quinn K.
        • et al.
        Research domain criteria (RDoC): Toward a new classification framework for research on mental disorders.
        Am J Psychiatry. 2010; 167: 748-751
        • Hyman S.E.
        The diagnosis of mental disorders: The problem of reification.
        Annu Rev Clin Psychol. 2010; 6: 155-179
        • Gillan C.M.
        • Kosinski M.
        • Whelan R.
        • Phelps E.A.
        • Daw N.D.
        Characterizing a psychiatric symptom dimension related to deficits in goal-directed control.
        Elife. 2016; 5e11305
        • Patzelt E.H.
        • Kool W.
        • Millner A.J.
        • Gershman S.J.
        Incentives boost model-based control across a range of severity on several psychiatric constructs.
        Biol Psychiatry. 2019; 85: 425-433
        • Amlung M.
        • Marsden E.
        • Holshausen K.
        • Morris V.
        • Patel H.
        • Vedelago L.
        • et al.
        Delay discounting as a transdiagnostic process in psychiatric disorders: A meta-analysis.
        JAMA Psychiatry. 2019; 76: 1176-1186
        • Gillan C.M.
        • Kalanthroff E.
        • Evans M.
        • Weingarden H.M.
        • Jacoby R.J.
        • Gershkovich M.
        • et al.
        Comparison of the association between goal-directed planning and self-reported compulsivity vs obsessive-compulsive disorder diagnosis [published correction appears in JAMA Psychiatry 2020; 77:10].
        JAMA Psychiatry. 2020; 77: 77-85
        • Drysdale A.T.
        • Grosenick L.
        • Downar J.
        • Dunlop K.
        • Mansouri F.
        • Meng Y.
        • et al.
        Resting-state connectivity biomarkers define neurophysiological subtypes of depression [published correction appears in Nat Med 2017; 23:264].
        Nat Med. 2017; 23: 28-38
        • Pizzagalli D.A.
        • Smoski M.
        • Ang Y.S.
        • Whitton A.E.
        • Sanacora G.
        • Mathew S.J.
        • et al.
        Selective kappa-opioid antagonism ameliorates anhedonic behavior: Evidence from the Fast-fail Trial in Mood and Anxiety Spectrum Disorders (FAST-MAS) [published correction appears in Neuropsychopharmacology 2021; 46:2224].
        Neuropsychopharmacology. 2020; 45: 1656-1663
        • Price R.B.
        • Gillan C.M.
        • Hanlon C.
        • Ferrarelli F.
        • Kim T.
        • Karim H.T.
        • et al.
        Effect of experimental manipulation of the orbitofrontal cortex on short-term markers of compulsive behavior: A theta burst stimulation study.
        Am J Psychiatry. 2021; 178: 459-468
        • Eisenberg I.W.
        • Bissett P.G.
        • Zeynep Enkavi A.
        • Li J.
        • MacKinnon D.P.
        • Marsch L.A.
        • Poldrack R.A.
        Uncovering the structure of self-regulation through data-driven ontology discovery.
        Nat Commun. 2019; 10: 2319
        • First M.B.
        Current state of psychiatric nosology.
        in: Redish A.D. Gordon J.A. Computational Psychiatry. New Perspectives on Mental Illness. MIT Press, Cambridge, MA2016: 139-152
        • National Institute of Mental Health
        Research Domain Criteria (RDoC).
        (Available at:)
        • Kiluk B.D.
        • Yip S.W.
        • DeVito E.E.
        • Carroll K.M.
        • Sofuoglu M.
        Anhedonia as a key clinical feature in the maintenance and treatment of opioid use disorder.
        Clin Psychol Sci. 2019; 7: 1190-1206
        • Huys Q.J.
        • Pizzagalli D.A.
        • Bogdan R.
        • Dayan P.
        Mapping anhedonia onto reinforcement learning: A behavioural meta-analysis.
        Biol Mood Anxiety Disord. 2013; 3: 12
        • American Psychiatric Association
        Diagnostic and Statistical Manual of Mental Disorders (DSM-5).
        American Psychiatric Publishing, Washington, D.C.2013
        • Slattery D.A.
        • Cryan J.F.
        Modelling depression in animals: At the interface of reward and stress pathways.
        Psychopharmacology (Berl). 2017; 234: 1451-1465
        • Ward R.D.
        Methods for dissecting motivation and related psychological processes in rodents.
        Curr Top Behav Neurosci. 2016; 27: 451-470
        • Thomsen K.R.
        Measuring anhedonia: Impaired ability to pursue, experience, and learn about reward.
        Front Psychol. 2015; 6: 1409
        • Berridge K.C.
        Evolving concepts of emotion and motivation.
        Front Psychol. 2018; 9: 1647
        • Mathis M.W.
        • Mathis A.
        Deep learning tools for the measurement of animal behavior in neuroscience.
        Curr Opin Neurobiol. 2020; 60: 1-11
        • Sweis B.M.
        • Abram S.V.
        • Schmidt B.J.
        • Seeland K.D.
        • MacDonald 3rd, A.W.
        • Thomas M.J.
        • Redish A.D.
        Sensitivity to “sunk costs” in mice, rats, and humans.
        Science. 2018; 361: 178-181
        • Sweis B.M.
        • Thomas M.J.
        • Redish A.D.
        Beyond simple tests of value: Measuring addiction as a heterogeneous disease of computation-specific valuation processes.
        Learn Mem. 2018; 25: 501-512
        • Groman S.M.
        • Massi B.
        • Mathias S.R.
        • Lee D.
        • Taylor J.R.
        Model-free and model-based influences in addiction-related behaviors.
        Biol Psychiatry. 2019; 85: 936-945
        • Groman S.M.
        • Rich K.M.
        • Smith N.J.
        • Lee D.
        • Taylor J.R.
        Chronic exposure to methamphetamine disrupts reinforcement-based decision making in rats.
        Neuropsychopharmacology. 2018; 43: 770-780
        • Zhukovsky P.
        • Puaud M.
        • Jupp B.
        • Sala-Bayo J.
        • Alsiö J.
        • Xia J.
        • et al.
        Withdrawal from escalated cocaine self-administration impairs reversal learning by disrupting the effects of negative feedback on reward exploitation: A behavioral and computational analysis [published correction appears in Neuropsychopharmacology 2020; 45:714].
        Neuropsychopharmacology. 2019; 44: 2163-2173
        • Hitchcock P.F.
        • Fried E.I.
        • Frank M.J.
        Computational psychiatry needs time and context.
        Annu Rev Psychol. 2022; 73: 243-270
        • Colaizzi J.M.
        • Flagel S.B.
        • Joyner M.A.
        • Gearhardt A.N.
        • Stewart J.L.
        • Paulus M.P.
        Mapping sign-tracking and goal-tracking onto human behaviors.
        Neurosci Biobehav Rev. 2020; 111: 84-94
        • Joyner M.A.
        • Gearhardt A.N.
        • Flagel S.B.
        A translational model to assess sign-tracking and goal-tracking behavior in children.
        Neuropsychopharmacology. 2018; 43: 228-229
        • Eckstein M.K.
        • Wilbrecht L.
        • Collins A.G.E.
        What do reinforcement learning models measure? Interpreting model parameters in cognition and neuroscience.
        Curr Opin Behav Sci. 2021; 41: 128-137
        • Gueguen M.C.M.
        • Schweitzer E.M.
        • Konova A.B.
        Computational theory-driven studies of reinforcement learning and decision-making in addiction: What have we learned?.
        Curr Opin Behav Sci. 2021; 38: 40-48
        • van Borkulo C.
        • Boschloo L.
        • Borsboom D.
        • Penninx B.W.J.H.
        • Waldorp L.J.
        • Schoevers R.A.
        Association of symptom network structure with the course of depression [published correction appears in JAMA Psychiatry 2016; 73:412].
        JAMA Psychiatry. 2015; 72: 1219-1226
        • Chen H.
        • Nebe S.
        • Mojtahedzadeh N.
        • Kuitunen-Paul S.
        • Garbusow M.
        • Schad D.J.
        • et al.
        Susceptibility to interference between Pavlovian and instrumental control is associated with early hazardous alcohol use.
        Addict Biol. 2021; 26e12983
        • Wu W.
        • Zhang Y.
        • Jiang J.
        • Lucas M.V.
        • Fonzo G.A.
        • Rolle C.E.
        • et al.
        An electroencephalographic signature predicts antidepressant response in major depression.
        Nat Biotechnol. 2020; 38: 439-447
        • Yip S.W.
        • Scheinost D.
        • Potenza M.N.
        • Carroll K.M.
        Connectome-based prediction of cocaine abstinence.
        Am J Psychiatry. 2019; 176: 156-164
        • Lichenstein S.D.
        • Scheinost D.
        • Potenza M.N.
        • Carroll K.M.
        • Yip S.W.
        Dissociable neural substrates of opioid and cocaine use identified via connectome-based modelling.
        Mol Psychiatry. 2021; 26: 4383-4393
        • Budde M.
        • Anderson-Schmidt H.
        • Gade K.
        • Reich-Erkelenz D.
        • Adorjan K.
        • Kalman J.L.
        • et al.
        A longitudinal approach to biological psychiatric research: The PsyCourse study.
        Am J Med Genet B Neuropsychiatr Genet. 2019; 180: 89-102
        • Coppersmith D.D.L.
        • Dempsey W.
        • Kleiman E.
        • Bentley K.
        • Murphy S.
        • Nock M.
        Just-in-time adaptive interventions for suicide prevention: Promise, challenges, and future directions.
        PsyArXiv. 2021; https://doi.org/10.31234/osf.io/eg9fx
        • Nahum-Shani I.
        • Smith S.N.
        • Spring B.J.
        • Collins L.M.
        • Witkiewitz K.
        • Tewari A.
        • Murphy S.A.
        Just-in-time adaptive interventions (JITAIs) in mobile health: Key components and design principles for ongoing health behavior support.
        Ann Behav Med. 2018; 52: 446-462
        • Carpenter S.M.
        • Menictas M.
        • Nahum-Shani I.
        • Wetter D.W.
        • Murphy S.A.
        Developments in mobile health just-in-time adaptive interventions for addiction science.
        Curr Addict Rep. 2020; 7: 280-290
        • Hébert E.T.
        • Ra C.K.
        • Alexander A.C.
        • Helt A.
        • Moisiuc R.
        • Kendzor D.E.
        • et al.
        A mobile just-in-time adaptive intervention for smoking cessation: Pilot randomized controlled trial.
        J Med Internet Res. 2020; 22e16907
        • Nahum-Shani I.
        • Hekler E.B.
        • Spruijt-Metz D.
        Building health behavior models to guide the development of just-in-time adaptive interventions: A pragmatic framework.
        Health Psychol. 2015; 34S: 1209-1219
        • Kleiman E.M.
        • Nock M.K.
        Real-time assessment of suicidal thoughts and behaviors.
        Curr Opin Psychol. 2018; 22: 33-37
        • Ebner-Priemer U.W.
        • Trull T.J.
        Ecological momentary assessment of mood disorders and mood dysregulation.
        Psychol Assess. 2009; 21: 463-475
        • de Girolamo G.
        • Barattieri di San Pietro C.
        • Bulgari V.
        • Dagani J.
        • Ferrari C.
        • Hotopf M.
        • et al.
        The acceptability of real-time health monitoring among community participants with depression: A systematic review and meta-analysis of the literature.
        Depress Anxiety. 2020; 37: 885-897
        • Businelle M.S.
        • Hébert E.T.
        • Kendzor D.E.
        Introduction to the special issue on use of mobile technology for real-time assessment and treatment of substance-use disorders.
        Am J Drug Alcohol Abuse. 2018; 44: 499-501
        • Laumann T.O.
        • Gordon E.M.
        • Adeyemo B.
        • Snyder A.Z.
        • Joo S.J.
        • Chen M.Y.
        • et al.
        Functional system and areal organization of a highly sampled individual human brain.
        Neuron. 2015; 87: 657-670
        • Poldrack R.A.
        • Laumann T.O.
        • Koyejo O.
        • Gregory B.
        • Hover A.
        • Chen M.Y.
        • et al.
        Long-term neural and physiological phenotyping of a single human.
        Nat Commun. 2015; 6: 8885
        • Naselaris T.
        • Allen E.
        • Kay K.
        Extensive sampling for complete models of individual brains.
        Curr Opin Behav Sci. 2021; 40: 45-51
        • Yang Z.
        • Lewis L.D.
        Imaging the temporal dynamics of brain states with highly sampled fMRI.
        Curr Opin Behav Sci. 2021; 40: 87-95
        • Newbold D.J.
        • Dosenbach N.U.F.
        Tracking plasticity of individual human brains.
        Curr Opin Behav Sci. 2021; 40: 161-168
        • Moutoussis M.
        • Garzón B.
        • Neufeld S.
        • Bach D.R.
        • Rigoli F.
        • Goodyer I.
        • et al.
        Decision-making ability, psychopathology, and brain connectivity.
        Neuron. 2021; 109: 2025-2040.e7
        • Ahn W.Y.
        • Vasilev G.
        • Lee S.H.
        • Busemeyer J.R.
        • Kruschke J.K.
        • Bechara A.
        • Vassileva J.
        Decision-making in stimulant and opiate addicts in protracted abstinence: Evidence from computational modeling with pure users.
        Front Psychol. 2014; 5: 849
        • Charpentier C.J.
        • Aylward J.
        • Roiser J.P.
        • Robinson O.J.
        Enhanced risk aversion, but not loss aversion, in unmedicated pathological anxiety.
        Biol Psychiatry. 2017; 81: 1014-1022
        • Wise T.
        • Dolan R.J.
        Associations between aversive learning processes and transdiagnostic psychiatric symptoms in a general population sample.
        Nat Commun. 2020; 11: 4179
        • Rouault M.
        • Seow T.
        • Gillan C.M.
        • Fleming S.M.
        Psychiatric symptom dimensions are associated with dissociable shifts in metacognition but not task performance.
        Biol Psychiatry. 2018; 84: 443-451
        • Baker S.C.
        • Konova A.B.
        • Daw N.D.
        • Horga G.
        A distinct inferential mechanism for delusions in schizophrenia.
        Brain. 2019; 142: 1797-1812
        • Maxim L.D.
        • Niebo R.
        • Utell M.J.
        Screening tests: A review with examples [published correction appears in Inhal Toxicol 2019; 31:298].
        Inhal Toxicol. 2014; 26: 811-828
        • Wilson J.M.G.
        • Jungner G.
        Principles and Practice of Screening for Disease.
        World Health Organization, Geneva1968
        • Ferrante di Ruffano L.
        • Hyde C.J.
        • McCaffery K.J.
        • Bossuyt P.M.M.
        • Deeks J.J.
        Assessing the value of diagnostic tests: A framework for designing and evaluating trials.
        BMJ. 2012; 344: e686
      1. Chang S.M. Matchar D.B. Smetana G.W. Umscheid C.A. Methods Guide for Medical Test Reviews. Agency for Healthcare Research and Quality, Rockville, MD2012
        • Rector T.S.
        • Taylor B.C.
        • Wilt T.J.
        Chapter 12: Systematic review of prognostic tests.
        J Gen Intern Med. 2012; 27: S94-S101
        • Center for Behavioral Health Statistics and Quality
        Behavioral Health Trends in the United States: Results From the 2014 National Survey on Drug Use and Health. Substance Abuse and Mental Health Services Administration, Rockville, MD2015
        • Yuksel S.E.
        • Wilson J.N.
        • Gader P.D.
        Twenty years of mixture of experts.
        IEEE Trans Neural Netw Learn Syst. 2012; 23: 1177-1193
        • Jacobs R.A.
        • Jordan M.I.
        • Nowlan S.J.
        • Hinton G.E.
        Adaptive mixtures of local experts.
        Neural Comput. 1991; 3: 79-87
        • Downar J.
        • Bhatt M.
        • Montague P.R.
        Neural correlates of effective learning in experienced medical decision-makers.
        PLoS One. 2011; 6e27768
        • Müller G.
        • Pfinder M.
        • Schmahl C.
        • Bohus M.
        • Lyssenko L.
        Cost-effectiveness of a mindfulness-based mental health promotion program: Economic evaluation of a nonrandomized controlled trial with propensity score matching.
        BMC Public Health. 2019; 19: 1309