Enhancing the Personalization of Psychotherapy With Dynamic Assessment and Modeling

Abstract

Clinicians have long recognized the importance of tailoring psychotherapy interventions to the needs and characteristics of the individual patient. However, traditional approaches to clinical assessment, service delivery, and intervention research have not been conducive to such personalization. Contrary to traditional nomothetic approaches, idiographic assessment and modeling of intraindividual dynamic processes holds tremendous promise for tailoring the implementation of psychotherapy to the individual patient. In this article, we (a) present an argument for assessing person-specific dynamics, (b) provide a detailed description of a method that harnesses person-specific dynamic assessment and modeling for use in routine psychotherapy, (c) present exemplar clinical cases illustrating these methods, and (d) discuss how these methods can be translated into routine clinical assessment and psychotherapy.

Keywords

person-specific methods dynamic modeling personalized medicine psychotherapy

Despite widespread agreement in the field that psychological interventions must be tailored to the individual patient, traditional approaches to clinical assessment, treatment delivery, and applied research have been limited in their attempts at fostering truly personalized case formulations and treatment plans. There has been increasing recognition for the need to develop innovative assessment methods that can inform more person-centered and responsive care. The current article presents a data-driven approach for personalized assessment, intervention selection, and targeted, modular treatment delivery. We describe a personalized modular treatment trial in which our research group has developed an assessment paradigm for identifying person-specific dimensions of psychopathology (Fisher, 2015) that cut across existing diagnostic categories to select optimal treatment protocols on a person-by-person basis. We highlight how the dynamic assessment approach can (a) enhance personalization, (b) accommodate comorbidity, (c) examine both within- and between-time covariation among intraindividual symptoms, and (d) reveal latent pathologies. We detail these strengths below and provide exemplar clinical cases for further illustration.

Personalization

Over the past decade, a compelling case has been made for the personalization of psychological science (Fisher & Bosley, 2015). From a methodological perspective, several authors have questioned the generalizability of nomothetic research findings to individuals (Fisher, 2015; Fisher, Newman, & Molenaar, 2011; Hamaker, Dolan, & Molenaar, 2005), and Molenaar (2004) has gone so far as to say that we should have no reason to expect any meaningful relationship between intra- and interindividual variation. That is, the variances and covariances among a set of variables at the individual level (varying over time) are likely to be statistically unrelated to the variances and covariances at the group level (varying across individuals). Such an assertion has important implications for the scientific study of psychological assessment and psychotherapy outcome. Best practice treatment decisions in psychiatry and psychology are overwhelmingly based on randomized controlled trials, experimental designs intended to reveal relatively macroscopic differences between large groups (i.e., treatment conditions). Moreover, the post hoc moderator analyses typically employed to identify characteristics of optimal subpopulations within these studies have often suffered from statistical bias due to a lack of random assignment and poor replication. Most, if not all, randomized controlled trials are powered only to examine a between group difference on a primary outcome measure, and are therefore underpowered to rigorously examine between-patient moderators. Thus, many have called for a renewed emphasis on idiographic research (Barlow & Nock, 2009; Nesselroade, Gerstorf, Hardy, & Ram, 2007) and efforts to personalize psychosocial interventions (DeRubeis et al., 2014; Fisher, 2015).

The movement toward personalization generates a greater appreciation for the extraordinary complexity of the individual patient and psychotherapy. All existing evidence-based treatments involve multiple components (Nathan & Gorman, 2015), hence the phrase “treatment packages,” with little systematic guidance regarding optimal sequencing or which treatment components are likely to be more or less important for a given patient. Dismantling or additive component studies are primarily motivated by an interest in identifying the specific active ingredients of a multicomponent treatment. However, neither the selection of components nor the sequencing of components is based on a personalized assessment and data are still aggregated across patients and examined at the group level. Perhaps not surprisingly, a recent meta-analysis of component treatment studies found no significant differences between full-treatment packages and dismantled treatments; however, there was a modest difference favoring treatments with specific additive components when compared with the standard treatment package (Bell, Marcus, & Goodlad, 2013). Unlike dismantling conditions, the integration of additive components is more likely to be based on putative patient-level moderators (see, e.g., Newman et al., 2011). This is a closer step toward personalization, in theory, yet is ultimately unsatisfactory because treatment selection and delivery is still being determined at the group level, there may still be significant interindividual response variability in the covariation between specific components and their sequencing, and there may be a confounding method effect when a component is always introduced at the same point in a sequence across groups of patients (Rehm, 2009). A particular advantage of a person-specific dynamic assessment approach is its potential to elucidate intraindividual mechanisms of psychopathology that are theoretically and empirically linked with specific treatment components that can be maximally tailored to this patient at this time.

Actionable, Dimensional Classification

A number of studies have provided evidence supporting the dimensional structure of mental illness and provided models for dimensional classification (Brown & Barlow, 2009; Wright et al., 2013). However, despite the importance of this work for understanding the taxonomic structure of psychopathology, existing dimensional systems have shown limited utility beyond classification or categorization. In order to move researchers and clinicians closer to a practical, applicable dimensional model, it is necessary to identify dimensions that provide actionable information about prescribed interventions. The current person-specific dynamic assessment approach achieves this by identifying the latent dimensions of psychopathology, person by person, as well as the dynamic organization of those dimensions in time. This dynamic assessment approach follows the framework recently outlined by Fisher (2015). We describe below a set of methodologies in which we (a) reduce broad, complex psychopathological symptom profiles to person-relevant dimensions, and (b) assess the presentation of these dimensions over time in order to (c) elucidate the underlying predictive dynamics among symptoms and behaviors. These methods utilize intensive repeated measurements of within-individual data (i.e., ecological momentary assessment); factor analyses of within-individual data for assessing latent structures in dynamic processes over time; and dynamic factor modeling (Molenaar, 1985), or the examination of underlying predictive relationships between symptoms over time.

Comorbidity, Co-Occurrence, and Their Underlying Dimensions

The problem of comorbidity is not limited to the Diagnostic and Statistical Manual of Mental Disorders (DSM; American Psychiatric Association, 2013), nor even necessarily to categorical taxonomies (though the term implies discrete disease states). For instance, the strong, positive correlation between depression and anxiety is a well-known and well-established phenomenon shown to manifest across clinical and nonclinical populations and between both categorical and dimensional representations (Brown, Chorpita, & Barlow, 1998; Clark & Watson, 1991; Clark, Watson, & Mineka, 1994; Costello, 1976; Dobson, 1985a, Dobson, 1985b; Gotlib, 1984; Mineka, Watson, & Clark, 1998; Watson & Clark, 1984; Watson, Clark, & Carey, 1988). It has been argued that the covariance between mood and anxiety—at both dimensional and categorical levels of analysis—is due to underlying hierarchical constructs such as negative affectivity and neuroticism (cf. Brown et al., 1998). However, the degree to which a given comorbid presentation of anxiety and depression is due to shared versus independent variance remains an empirical question, and one whose answer may differ from person to person.

Extant, previously labeled, Axis I diagnoses are characterized by an array of temporally dependent symptoms. In the mood and anxiety disorders, symptoms such as perseverative thought and depressed mood co-occur and tend to resolve together in treatment. Of the seven diagnostic criteria for generalized anxiety disorder (GAD) and nine diagnostic criteria for major depressive disorder (MDD), four are shared symptoms: fatigue, sleep disturbance, psychomotor agitation (e.g., restless and keyed up), and difficulty concentrating. Despite the large degree of shared variance between them, there is a wide array of possible symptom presentations in MDD and GAD as currently defined by the DSM. Thus, there is a pressing need to separate or filter the signal from the noise by identifying the meaningful expressions of pathology for each individual.

The dynamic assessment approach recently proposed by Fisher (2015) endeavors to identify the latent dimensions underlying psychopathology, whether across one or multiple putative diagnostic categories. This approach seeks to reveal latent dimensions both in terms of the factor structure of latent variable models as well as latent pathological dimensions that may otherwise go undetected (i.e., latent in both the statistical and metaphysical sense).

Symptom Covariation in Time

Relatively little attention has been paid to—and thus, very little is known about—the dynamic relationships between symptoms over time. Both psychological assessments and psychosocial interventions make implicit assumptions of causal equivalence across symptom domains, making little effort to elucidate the predominance or causal hierarchy among symptoms. Thus, it is possible that the relatively modest response rates across large-scale outcome studies may result (at least in part) from the standardized treatment of dynamically heterogeneous disorders. The dynamic assessment approach can potentially increase therapeutic efficacy and treatment response by identifying and incorporating such heterogeneity into treatment through personalized designs.

In addition to the ability to parse core components of psychopathology from peripheral phenomena, a particular strength of the person-specific dynamic assessment approach is the ability to examine the dynamics between latent dimensions both as they co-occur in a given moment, as well as how they predict each other from moment to moment within a single individual. These distinctions, between contemporaneous and time-lagged relationships, and between the group and the individual, are central to the dynamic assessment approach, for which the broad goal is to understand psychopathology as a set of dynamic processes—in which thoughts, feelings, and actions are idiosyncratically organized within each individual. We believe that understanding the organization of these phenomena within an individual, and the manner in which they unfold in time, can lead to the delivery of interventions that target the internal structure of the pathologic syndrome. We offer as an example the aforementioned correlation between anxiety and depression.

As noted above, decades of research has substantiated a correlation between anxiety and depression of roughly .50 to .70 (e.g., Clark & Watson, 1991). This relationship has been established across categories and dimensions, within healthy and clinical populations (Brown, 2007; Kessler, Chiu, Demler, & Walters, 2005; Lovibond, 1998). However, overall correlations at the group level may not generalize to individuals, and even if a correlation exists across individuals, it may not exist across time for a given individual (Molenaar, 2004). At the group level, the correlation between anxiety and depression demonstrates that, in a given sample, individuals relatively higher in rank-ordered levels of depression will be (roughly) equivalently rank ordered on levels of anxiety. The strength of the relationship is dependent on the relative position of depression and anxiety scores across a group of individuals. However, it says nothing of the dynamic or mechanistic relationship between the two constructs as they manifest within an individual.

By collecting a large number of observations of depression and anxiety over time, we can assess the relationship between these constructs within a single individual—replacing the standard sample size requirement of large numbers of participants with an equivalent number of time points. This evaluation remains correlational, yet is reflective of the relative behavior of these constructs as they vary within—and are experienced by—the individual. Moreover, by utilizing time series methodologies such as the dynamic factor model, we can concurrently assess the contemporaneous and time-lagged relationships among the constructs. That is, the degree to which they vary together at a given moment in time and the degree to which one may drive the variation in the other from moment to moment. This distinction is paramount. As we describe above, it is believed that the strong correlation between depression and anxiety is a result of an underlying source of general negative affect. Thus, individuals within a sample with higher levels of overall distress will be more likely to exhibit higher depression and higher anxiety. However, this influence is largely obviated in within-person designs. Later, we provide an example of an individual for whom a moderate positive correlation (r = .39) was observed contemporaneously, while a negative correlation was observed across time (β = −.28).

Dynamic Assessment Methodology

This dynamic assessment approach consists of a set of person-specific methodologies for data collection, analysis, and interpretation. In the following, we detail how data are collected, the specific methods of analysis, and procedures for utilizing these analyses for the construction and application of personalized modular therapies.

Study Design

The following methods and patient-participant exemplars were taken from an ongoing study in the first author’s lab at the University of California, Berkeley. Individuals with symptomatic experiences consistent with GAD and MDD were directed by flyers, referrals, and Internet advertisements to contact the Idiographic Dynamics Lab at University of California, Berkeley. Potential participants were screened (after obtaining verbal consent) through a brief telephone interview, including a review of diagnostic suitability and current usage of medications, and administration of the Mini-International Neuropsychiatric Interview (Sheehan et al., 1998). Based on this preliminary information, eligible participants who met initial inclusion criteria were invited for an in-person structured clinical interview. The Anxiety Disorders Interview Schedule (ADIS) for DSM-5 Brown & Barlow, 2014) was administered by graduate students in clinical psychology. Inclusion criteria were primary diagnosis of GAD and/or MDD, age of 18 to 65 years, and a web-enabled mobile phone. Exclusion criteria were any history of psychosis or mania, concurrent treatment or cognitive-behavioral treatment within the past 12 months, and PRN medication. Following enrollment, participants completed smartphone-enabled, web-based surveys four times per day for a minimum of 30 days. A minimum response rate of 80% was required. Following the data analytic procedures described below, a personalized modular therapy of variable length was provided by postdoctoral therapists. The ADIS-5 was readministered following the conclusion of therapy.

Intensive Repeated Measures Methodology

Participants were surveyed four times a day for 30 days via a web-based survey system that sends e-mails to participants’ personal smartphones that are received as text messages. Participants were asked to provide their standard waking time and the time window specified by the participant was split into four equal blocks of 3 hours. Surveys were then conducted at a random time within each block, with the additional constraint that surveys should be given at least 30 minutes apart. At each sampling occasion, participants were prompted to think about the period of time since the last survey. In addition to the extant DSM-5 symptom criteria for GAD and MDD, surveys included three behavioral symptoms: avoided activities, procrastinated, and sought reassurance. Participants rated their experience of each symptom and behavior on a 0-100-visual analog slider, with anchors of not at all and as much as possible anchored at the 0 and 100 positions, respectively. Each survey prompt, whether completed or not, was time-stamped to facilitate the calculation of cumulative time throughout the survey completion period.

P-Technique Factor Analysis

The collection of intensive repeated measurements, four times per day for 30 days, yields a rich, 13 × 120 dimensioned multivariate time series. For each individual, this time series was subjected to a person-specific factor analysis (i.e., P-technique; Cattell, Cattell, & Rhymer, 1947) to identify latent dimensions across symptoms of anxiety and depression. Consistent with Fisher (2015), a two-step approach was taken whereby we delineated the structure of individual syndromes via exploratory factor analysis and confirmed this structure via confirmatory factor analysis. The confirmatory step eschews factor loadings less than .30 and thus provides a more rigorous test of the factor structure. Moreover, the Lagrange multiplier test provided by LISREL illuminates any points of model strain, in the event of cross-loadings or misspecified loadings. Although in nomothetic research either of these steps may be considered a risk for “overfitting,” we consider these the inverse—a maximally conservative approach to determining the idiographic factor structure. For exploratory models, an iterative approach for determining the number of factors was employed whereby a one-factor model was initially tested and assessed for model fit, followed by two-, three-, and four-factor models (when indicated). A final model was retained when an acceptable fit is indicated by the (a) chi-square goodness-of-fit statistic, (b) root mean square error of approximation and (c) standardized root mean square residual (for a review of these criteria and their respective optimal cutoffs, see Hu & Bentler, 1999).

Dynamic Factor Modeling

The dynamic factor model (Molenaar, 1985) is a vector-autoregressive methodology that utilizes a structural equation model framework to assess contemporaneous correlations and time-lagged regressions in multivariate time series. Once a final confirmatory model was selected, factor scores were generated from the input data by multiplying the raw data matrix by a weighting matrix composed of the factor loadings from the confirmatory solution. Because the data were unevenly sampled (with intervals of ~4, ~4, ~4, and ~12 hours) a cubic spline interpolation was applied to the factor-scored time series, in order to produce a time series with even, 6-hour intervals. The interpolated time series were each duplicated and lagged by a single observation, yielding a data frame with a set of time-lagged factors (time t − 1) and time-forward factors (time t), and FIML estimation was used to analyze the raw data in LISREL. For each individual, an initial model was run with all contemporaneous factor correlations and autoregressions. The Lagrange multiplier test was employed to detect the cross-lagged regression relationships in the data (Chou & Bentler, 1990). Finally, nonsignificant autoregressions were fixed to zero after all cross-lagged parameters had been added. As such, this approach uses dynamic factor models to examine dynamic (e.g., moment-to-moment) relationships within and between constructs in time, in order to identify predominant symptomatic dimensions.

Modular Therapy

As noted above, most existing evidence-based treatments are DSM diagnosis-specific treatment packages that include multiple components to be delivered in a standardized sequence. A potential paradigm shift has been marked by the development of transdiagnostic and modular treatments that have the potential benefit of (a) targeting common underlying dimensions of mood and anxiety disorders and (b) enhancing personalization by promoting greater flexibility in both the sequencing and duration of specific treatment components. Accumulating research is demonstrating the utility of transdiagnostic models and protocols (Barlow, Sauer-Zavala, Carl, Bullis, & Ellard, 2013; Brown & Barlow, 2009). We chose to utilize the Unified Protocol for Transdiagnostic Treatment of Emotional Disorders (UP; Barlow et al., 2011) because of its modular design and demonstrated efficacy across anxiety and mood disorders (e.g., Boswell et al., 2013).

The UP is a paradigm-shifting intervention, yet several potential limitations still exist in current transdiagnostic approaches to treatment. Chief among these are the putative uniformity of psychopathology across individuals and the standardized delivery of treatment to date. Although the UP is intended for flexible application and module sequencing, existing trials have focused on a standardized module sequence. Standardization of treatment content and delivery is essential to gold standard methods for establishing efficacy in traditional clinical trials; however, this can also result in treatments that are simultaneously incomplete and yet too diffuse. In our view, the full potential for treatments such as the UP, with respect to clinical impact and efficiency, will be realized when they are used to (a) target active psychopathological dimensions within individuals and (b) deliver core intervention components in a hierarchical sequence.

Therefore, we have utilized the UP as a “menu” of treatment modules, each designed to treat a targeted subset of symptom domains. Only those modules related to identified pathologic dimensions within each individual are selected for treatment delivery, allowing therapists to eschew unnecessary or irrelevant therapeutic elements for more efficient, personalized interventions. In addition, the order of modules is determined by the dynamic relationships indicated by person-specific dynamic factor models. That is, interventions for symptoms shown to drive the behavior of other symptoms are preferentially delivered earlier in therapy. Our group has developed a selection algorithm—the dynamic assessment treatment algorithm (DATA)—for selecting and ordering treatment modules based on the hierarchy of symptoms and behaviors.

Treatment Selection Algorithm

Our group has employed a two-stage approach to the development of a treatment selection algorithm. As described by Fernandez, Fisher, and Chi (2016), the first phase of development utilized an expert clinical panel, consisting of the first author, a postdoc in clinical psychology, and a practicing clinician from the local community. Moreover, the panel was in regular consultation with the second author. For each of the first 20 participants (and 13 treatment completers), the panel met to review the results of the factor analyses and dynamic models and formulated personalized treatment plans contingent on symptom predominance. Both symptom hierarchy and module selection were based on the observed explanatory power of identified factors within-time (factor analysis) and across-time (dynamic modeling). The designation of within-time and across-time was considered paramount to treatment planning. Factors accounting for the greatest variance were considered to best define the structure and organization of the presenting syndrome. Among the dynamic interrelationships, those factors exhibiting cross-lagged predictive effects on other factors were thought to potentiate downstream effects from early gains, and were thus given preferential consideration. However, as Fernandez et al. (2016) describe, this approach was overly reliant on the thematic interpretation of each factor solution and the application of semantic labels. Thus, indicators with lesser factor loadings might be overlooked or undervalued, to the degree that they did not align with the thematic label (e.g., “depression,” “anxiety,” “avoidance”) applied to the factor. We determined that it was necessary to more objectively quantify the contribution of all presenting symptoms and behaviors, and to do so in a way that was fully digitized and systematic.

Utilizing the data from the first 20 participants for pilot testing, DATA was designed to quantify the contribution of each individual symptom or behavior—as a function of the factor to which each symptom belonged. A set of weighting equations were devised (and embedded into a Microsoft Excel–based calculator) in order to (a) calculate the predominance among factors within time as reflected by the percent variance accounted for among presenting symptoms and behaviors, (b) calculate the predominance of factors across time as reflected by the percentage of predictive (lagged regression) variance accounted for by each factor, and (c) assign an item score for each symptom or behavior based on the relative strength of association (i.e., standardized factor loading) between the item and its associated factor. Finally, it was observed that our initial approach to treatment selection had been exclusively driven by variance–covariance relationships, ignoring the contribution of symptom severity (i.e., mean level). Thus, the final step (d) weighted the item score by the mean level over the pretherapy period. Taken together, DATA assigns a factor score to each factor derived from Steps a and b and then an item score to each item based on Steps c and d. Module selection and ordering is then based on a logical matrix matching modules to items—with the highest scoring module delivered first, followed by successively lower scoring modules. A more complete description of DATA is outside of the bounds of the current article and we refer readers to Fernandez et al. (2016) for further details.

Two Clinical Exemplars

The exemplars to follow were the fourth and seventh patients to participate in the ongoing trial, and their treatment plans were derived by the clinical panel. As described below, “Peter” was provided Modules 2, 3, 5, 4, 7, and 8 and “Mary” was provided Modules 2, 3, 4, 5, and 8. In contrast to this, DATA selected Modules 2, 4, 5, 3, 7, and 8 and Modules 2, 3, 5, 4, and 8, respectively, for these participants. For Peter, “worried” exhibited the highest mean level and a moderate loading on the second highest rated factor. In addition, “hopeless” and “worthless and guilty” (the other two matches for Module 4, cognitive reappraisal) exhibited strong loadings on the highest rated factor, in addition to elevated mean levels. Thus, the need for a module that was deemphasized by the clinical team was better illuminated by the finer granularity of the DATA approach. Importantly, although the modules selected for each participant were highly overlapping, we believe that module ordering is a vital element of effective personalization. In addition to the potential for the downstream effects indicated by the dynamic models, components delivered earlier in therapy have potentially greater saturation, as these interventions have the opportunity to permeate throughout the therapy period and participants have a greater amount of time to practice newly acquired skills within the therapeutic context.

Example 1, Peter

Peter was a 23-year-old male presenting with a primary diagnosis of MDD, with comorbid GAD and social anxiety disorder. P-technique factor analysis revealed three factors accounting for 50% of the overall symptom variance. Table 1 provides the factor loadings for Peter’s P-technique model. The three factors were labeled depression (22% of the variance), anxiety (18% of the variance), and avoidance (10% of the variance).

Table 1.

Participants’ P-Technique Factor Solution.

	Peter			Mary
	Depression (22%)	Anxiety (18%)	Avoidance (10%)	Depression (21%)	Anxiety (18%)
Irritable	—	.65	—	.50	—
Restless	—	.52	—	—	.37
Worried	—	.54	—	—	.66
Worthless/guilty	.78	—	—	—	.55
Loss of interest or pleasure	.63	—	.30	.87	—
Hopeless	.65	—	—	.38	.35
Down/depressed	.95	—	—	.51	.36
Fatigue	—	—	.38	—	.49
Muscle tension	—	.69	—	.56	—
Difficulty concentrating	—	.60	—	—	.48
Avoid activities	—	—	.66	.59	—
Reassurance seeking	.41	.33	−.30	—	—
Procrastination	—	—	.55	—	.43

Note. Values in parentheses reflect the percentage of the total variance accounted for by each factor.

As we describe above, the dynamic factor model facilitates the examination of both within- and across-time covariation. The results of the dynamic factor model for Peter (Figure 1) revealed an important consequence of this approach. Whereas the contemporaneous correlation between anxiety and depression of r = .39 is consistent with the nomothetic research literature, the time-lagged regression parameter for the effect of depression at time (t − 1) on anxiety at time (t) is inconsistent with the well-established and methodologically robust correlation. That is, although the co-occurrence of these constructs is positively correlated at any given moment, their dynamic relationship is such that as levels of depression increase in one moment, levels of anxiety are likely to decrease in successive moments. Again, this is a level of granularity and specificity that would be otherwise lost if the data were treated at the nomothetic level. Moreover, these data provide valuable diagnostic insight. Whereas the question of diagnostic primacy and therapeutic priority is a classic dilemma in the treatment of comorbid mood and anxiety disorders (Barlow, 2004), these data indicate a predictive hierarchy in which the effective treatment of depressive symptomatology is likely to have significant downstream effects on anxiety. Thus, not only is the current idiographic approach able to delineate contemporaneous versus time-lagged covariation but it does so in a way that points directly to underlying dynamics in individual symptomatology that provide prescriptive information for targeted treatment.

Figure 1.

Dynamic factor model for Peter.

Peter was provided with 11 weeks of psychotherapy: two sessions each of UP Modules 2 (psychoeducation and emotion monitoring), 3 (present-focused awareness training), and 5 (reducing emotion avoidance and promoting more adaptive action tendencies), a single session of Module 4 (cognitive reappraisal), and then two sessions each of Modules 7 (emotion exposure) and 8 (relapse prevention). Outcome data were strong for this case. Figure 2 presents the change during therapy on the self-reported, latent dimensions of anxiety and depression.¹ Peter exhibited a 7-point change on the Hamilton Rating Scale for Anxiety (Hamilton, 1959) from 13 points at pretreatment to 6 at posttreatment, and a 5-point shift on the Hamilton Rating Scale for Depression (Hamilton, 1960) from 13 points to 8 points. Using an SD pooled across pre- and posttherapy for the 13 current treatment completers in the study under description, these changes reflected Cohen’s ds of −0.92 and −1.50, respectively.

Figure 2.

Symptom response for Peter during therapy.

Example 2, Mary

Mary was a 20-year-old female presenting with a primary diagnosis of GAD, with comorbid social anxiety disorder. P-technique factor analysis revealed two factors accounting for 40% of the overall symptom variance. Table 1 provides the factor loadings for Mary’s P-technique model. The two factors were depression (21% of the variance) and anxiety (18% of the variance). It is of great interest to note that Mary did not present with current symptoms of depression. However, this individual did have a history of a single past episode of MDD (at age 18). Consistent with this history, a latent depression variable was identified for this individual and this factor accounted for a greater percentage of the overall variance than did symptoms of anxiety and the corresponding anxiety factor—despite the fact that this individual presented with two current anxiety disorders.

Therefore, the dynamic assessment approach can be said to articulate latent dimensions of psychopathology not only as they relate to latent variable modeling and the covariation among items in a measurement model but also insofar as these dimensions are present but otherwise unexpressed. The ADIS is a gold-standard semistructured interview. The interviewer—a graduate student in clinical science—assessed the participant for current MDD and, finding no evidence for a current diagnosis, assessed for past history and correctly distinguished a past history of MDD. Yet a clear and coherent factor for current depression was returned, driven primarily by anhedonia. This factor showed a moderate, positive correlation with anxiety (r = .46) and a relatively high degree of stability from moment to moment (autoregression = .63). Eschewing interventions for depressive symptoms would likely have greatly undermined this individual’s treatment. In fact, an examination of Mary’s outcome data reveals a Cohen’s d of −3.01 on the Hamilton Rating Scale for Depression (a change from 17 points to 7), with a 2-point increase from 9 to 11 points on the Hamilton Rating Scale for Anxiety (a Cohen’s d of 0.26). Mary was provided with 12 weeks of psychotherapy: two sessions of UP Modules 2 (psychoeducation and emotion monitoring) and 3 (present-focused awareness training), three sessions of Modules 4 (cognitive reappraisal) and 5 (reducing emotion avoidance and promoting more adaptive action tendencies), and two sessions of Module 8 (relapse prevention).

Discussion

At a fundamental level, person-specific dynamic assessment and modeling fosters actuarially driven clinical decision making and clinical responsiveness (Constantino, Boswell, Bernecker, & Castonguay, 2013). Furthermore, although we do not wish to perpetuate what we view as a false dichotomy between clinical assessment and treatment, the selection and timing of specific psychotherapeutic interventions should flow logically from the person-specific structures and dynamics elucidated in the baseline assessment period. Rather than follow a practice guideline that might indicate a particular evidence-based treatment package for a particular diagnosis, the targets of interest are more clearly defined, often transdiagnostic, mechanisms of pathology. The decision to implement a given clinical intervention is nearly always an integration of one’s foundational theoretical model, case conceptualization, and empirical evidence. With some exceptions (e.g., Boswell, Anderson, & Barlow, 2014), direct empirical evidence supporting the use of specific intervention X to specifically target mechanism Y is often lacking. Certainly, more clinically meaningful research is needed to increase the precision of our assessment tools and enhance the selection and implementation of targeted treatment modules.

Transdiagnostic approaches that attempt to target shared, underlying mechanisms of pathology offer particular promise and a logical starting place for this work, in our view. Module development in the UP (Barlow et al., 2011) progressed along two key lines: (a) identifying shared, underlying mechanisms of pathology in diverse emotional disorders and (b) identifying discrete interventions in the existing evidence-based treatment literature that have either direct or indirect empirical support for “moving” the target mechanism. For example, emotion intolerance appears to cut across diverse DSM disorder categories (Otto, 2008), and interoceptive exposure has been shown to decrease levels of such intolerance in a variety of disorders (Boswell et al., 2013; Zvolensky & Otto, 2007). Consequently, the UP includes a module that targets emotion-related interoceptive sensitivity and intolerance (Module 6). If the results from a patient’s person-specific factor analysis and dynamic factor model indicate the presence of an interoceptive intolerance factor that appears to be driving other symptoms (e.g., increases in worry, behavioral avoidance), then a clinician has solid ground to integrate interoceptive exposure in early treatment sessions.

Compared with constructs such as interoceptive sensitivity, other latent factors and dynamic assessment results may be much less straightforward. For example, although reductions in negative repetitive thought (rumination and worry) are often observed in cognitive behavioral therapies (e.g., Newby, Williams, & Andrews, 2014), there appear to be “multiple roads to Rome” when addressing this transdiagnostic factor (e.g., mindfulness, relaxation, exposure). Future research will need to inform the selection of specific intervention strategies for identified mechanisms in particular types of patients. For example, dynamic assessment and modeling may indicate that the “lynchpin” factor for a given patient is negative repetitive thought; however, in the absence of identified moderators, the choice of a particular intervention to target negative repetitive thought may be less clear. In instances when multiple potential strategies have garnered similar empirical support, patient preferences should play a critical role in the decision-making process (Norcross & Wampold, 2011).

Feasibility

When it comes to the feasibility of integrating these methods in routine clinical practice, there is a paradox to dynamic assessment and modeling. On the one hand, this approach relies on sophisticated methods and technology, and at least on the surface, appears to increase the complexity of clinical decision making. On the other hand, it distills person-specific structures and dynamics that ultimately reduce complexity, can be more closely linked to intervention strategies, and may result in more efficient treatments. Research indicates that the majority of mental health care patients have access to a smartphone and a favorable attitude toward using such technology as part of treatment (Torous, Friedman, & Keshavan, 2014). We believe that the described methods can be distilled in a way that will ultimately be feasible to integrate into routine clinical practice. Whenever clinical decisions are guided by actuarial data, there is the potential to make better informed decisions and reduce treatment complexity. For example, treatment components determined prospectively to be “inactive” or irrelevant for a given patient would not be used, thereby resulting in a more efficient treatment with fewer moving parts. Such targeted intervention is more likely to enhance treatment impact and cost-effectiveness.

Conclusion

In this article, we have outlined and described methods for integrating person-specific dynamic assessment in clinical research and practice. Clinicians have long recognized the importance of tailoring assessment and intervention delivery to the needs and characteristics of the individual patient. However, traditional approaches to clinical assessment, service delivery, and intervention research have not been conducive to such personalization. Person-specific dynamic assessment holds tremendous promise for adapting the implementation of psychotherapy to the individual. Work in this area is deservedly receiving more attention, and we believe that this will result in enhanced patient benefit.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

American Psychiatric Association. (2013). The Diagnostic and Statistical Manual of Mental Disorders–Fifth edition. Arlington, VA: Author.

Barlow

D. H.

(2004). Anxiety and its disorders: The nature and treatment of anxiety and panic: New York, NY: Guilford Press.

Barlow

D. H.

Farchione

T. J.

Fairholme

C. P.

Ellard

K. K.

Boisseau

C. L.

Allen

L. B.

Ehrenreich-May

(2011). Unified protocol for transdiagnostic treatment of emotional disorders: Therapist guide. New York, NY: Oxford University Press.

Barlow

D. H.

Nock

M. K.

(2009). Why can’t we be more idiographic in our research? Perspectives on Psychological Science, 4, 19-21.

Barlow

D. H.

Sauer-Zavala

Carl

J. R.

Bullis

J. R.

Ellard

K. K.

(2013). The nature, diagnosis, and treatment of neuroticism: Back to the future. Clinical Psychological Science, 2, 344-365. doi:10.1177/2167702613505532

Bell

E. C.

Marcus

D. K.

Goodlad

J. K.

(2013). Are the parts as good as the whole? A meta-analysis of component treatment studies. Journal of Consulting and Clinical Psychology, 81, 722-736. doi:10.1037/a0033004

Boswell

J. F.

Anderson

L. M.

Barlow

D. H.

(2014). An idiographic analysis of change processes in the unified transdiagnostic treatment of depression. Journal of Consulting and Clinical Psychology, 82, 1060-1071. doi:10.1037/a0037403

Boswell

J. F.

Farchione

T. J.

Sauer-Zavala

Murray

H. W.

Fortune

M. R.

Barlow

D. H.

(2013). Anxiety sensitivity and interoceptive exposure: A transdiagnostic construct and change strategy. Behavior Therapy, 44, 417-431.

Brown

T. A.

(2007). Temporal course and structural relationships among dimensions of temperament and DSM-IV anxiety and mood disorder constructs. Journal of Abnormal Psychology, 116, 313-328.

10.

Brown

T. A.

Barlow

D. H.

(2009). A proposal for a dimensional classification system based on the shared features of the DSM-IV anxiety and mood disorders: Implications for assessment and treatment. Psychological Assessment, 21, 256-271.

11.

Brown

T. A.

Barlow

D. H.

(2014). Anxiety and related disorders interview schedule for DSM-5: Clinician manual. New York, NY: Oxford University Press.

12.

Brown

T. A.

Chorpita

B. F.

Barlow

D. H.

(1998). Structural relationships among dimensions of the DSM-IV anxiety and mood disorders and dimensions of negative affect, positive affect, and autonomic arousal. Journal of Abnormal Psychology, 107, 179-192. doi:10.1037/0021-843X.107.2.179

13.

Cattell

Rhymer

R. M.

(1947). P-technique demonstrated in determining psychophysiological source traits in a normal individual. Psychometrika, 12, 267-288.

14.

Chou

C. P.

Bentler

P. M.

(1990). Model modification in covariance structure modeling: A comparison among likelihood ratio, lagrange multiplier, and Wald tests. Multivariate Behavioral Research, 25, 115-136. doi:10.1207/s15327906mbr2501_13

15.

Clark

L. A.

Watson

(1991). Tripartite model of anxiety and depression: Psychometric evidence and taxonomic implications. Journal of Abnormal Psychology, 100, 316-336.

16.

Clark

L. A.

Watson

Mineka

(1994). Temperament, personality, and the mood and anxiety disorders. Journal of Abnormal Psychology, 103, 103-116.

17.

Constantino

M. J.

Boswell

J. F.

Bernecker

S. L.

Castonguay

L. G.

(2013). Context-responsive psychotherapy integration as a framework for a unified clinical science: Conceptual and empirical considerations. Journal of Unified Psychotherapy and Clinical Science Volume, 2(1), 1-20.

18.

Costello

C. G.

(1976). Anxiety and depression: The adaptive emotions. Oxford, England: McGill-Queen’s University Press.

19.

DeRubeis

R. J.

Cohen

Z. D.

Forand

N. R.

Fournier

J. C.

Gelfand

L. A.

Lorenzo-Luaces

(2014). The Personalized Advantage Index: Translating research on prediction into individualized treatment recommendations. A demonstration. PLoS ONE, 9(1), e83875. doi:10.1371/journal.pone.0083875

20.

Dobson

K. S.

(1985a). An analysis of anxiety and depression scales. Journal of Personality Assessessment, 49, 522-527.

21.

Dobson

K. S.

(1985b). The relationship between anxiety and depression. Clinical Psychology Review, 5, 307-324.

22.

Fernandez

K. C.

Fisher

A. J.

Chi

(2016). Development and initial implementation of the Dynamic Assessment Treatment Algorithm (DATA). Manuscript submitted for publication.

23.

Fisher

A. J.

(2015). Toward a dynamic model of psychological assessment: Implications for personalized care. Journal of Consulting and Clinical Psychology, 83, 825-836. doi:10.1037/ccp0000026

24.

Fisher

A. J.

Bosley

H. G.

(2015). Personalized assessment and treatment of depression. Current Opinion in Psychology, 4, 67-74. doi:10.1016/j.copsyc.2015.03.031

25.

Fisher

A. J.

Newman

M. G.

Molenaar

P. C. M.

(2011). A quantitative method for the analysis of nomothetic relationships between idiographic structures: Dynamic patterns create attractor states for sustained posttreatment change. Journal of Consulting and Clinical Psychology, 79, 552-563. doi:10.1037/a0024069

26.

Gotlib

I. H.

(1984). Depression and general psychopathology in university students. Journal of Abnormal Psychology, 93, 19-30.

27.

Hamaker

E. L.

Dolan

C. V.

Molenaar

P. C. M.

(2005). Statistical modeling of the individual: Rationale and application of multivariate stationary time series analysis. Multivariate Behavioral Research, 40, 207-233.

28.

Hamilton

(1959). The assessment of anxiety states by rating. British Journal of Medical Psychology, 32, 50-55.

29.

Hamilton

(1960). A rating scale for depression. Journal of Neurology, Neurosurgery, and Psychiatry, 23, 56-62.

30.

L.-t.

Bentler

P. M.

(1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6, 1-55.

31.

Kessler

R. C.

Chiu

W. T.

Demler

Walters

E. E.

(2005). Prevalence, severity, and comorbidity of 12-month DSM-IV disorders in the national comorbidity survey replication. Archives of General Psychiatry, 62, 617-627.

32.

Lovibond

P. F.

(1998). Long-term stability of depression, anxiety, and stress syndromes. Journal of Abnormal Psychology, 107, 520-526.

33.

Mineka

Watson

Clark

L. A.

(1998). Comorbidity of anxiety and unipolar mood disorders. Annual Review of Psychology, 49, 377-412.

34.

Molenaar

P. C. M.

(1985). A dynamic factor model for the analysis of multivariate time series. Psychometrika, 50, 181-202.

35.

Molenaar

P. C. M.

(2004). A manifesto on psychology as idiographic science: Bringing the person back into scientific psychology, this time forever. Measurement: Interdisciplinary Research and Perspectives, 2, 201-218.

36.

Nathan

P. E.

Gorman

J. M.

(2015). A guide to treatments that work. New York, NY: Oxford University Press.

37.

Nesselroade

J. R.

Gerstorf

Hardy

S. A.

Ram

(2007). Focus article: Idiographic filters for psychological constructs. Measurement: Interdisciplinary Research and Perspectives, 5, 217-235. doi:10.1080/15366360701741807

38.

Newby

J. M.

Williams

A. D.

Andrews

(2014). Reductions in negative repetitive thinking and metacognitive beliefs during transdiagnostic internet cognitive behavioural therapy (iCBT) for mixed anxiety and depression. Behaviour Research and Therapy, 59, 52-60.

39.

Newman

M. G.

Castonguay

L. G.

Borkovec

T. D.

Fisher

A. J.

Boswell

J. F.

Szkodny

L. E.

Nordberg

S. S.

(2011). A randomized controlled trial of cognitive-behavioral therapy for generalized anxiety disorder with integrated techniques from emotion-focused and interpersonal therapies. Journal of Consulting and Clinical Psychology, 79, 171-181. doi:10.1037/a0022489

40.

Norcross

J. C.

Wampold

B. E.

(2011). What works for whom: Tailoring psychotherapy to the person. Journal of Clinical Psychology, 67, 127-132. doi:10.1002/jclp.20764

41.

Otto

M. W.

(2008). Anxiety sensitivity, emotional intolerance, and expansion of the application of interoceptive exposure: Commentary on the special issue. Journal of Cognitive Psychotherapy, 22, 379-384.

42.

Rehm

L. P.

(2009). Looking at the components of treatment for depression. Clinical Psychology: Science and Practice, 16, 412-415.

43.

Sheehan

D. V.

Lecrubier

Sheehan

K. H.

Amorim

Janavs

Weiller

. . . Dunbar

G. C.

(1998). The Mini-International Neuropsychiatric Interview (MINI): The development and validation of a structured diagnostic interview for DSM-IV and ICD-10. Journal of Clinical Psychiatry, 59(Suppl. 20), 20-23.

44.

Torous

Friedman

Keshavan

(2014). Smartphone ownership and interest in mobile applications to monitor symptoms of mental health conditions. JMIR mHealth and uHealth, 2(1), e2. doi:10.2196/mhealth.2994

45.

Watson

Clark

L. A.

(1984). Negative affectivity: The disposition to experience aversive emotional states. Psychological Bulletin, 96, 465-490.

46.

Watson

Clark

L. A.

Carey

(1988). Positive and negative affectivity and their relation to anxiety and depressive disorders. Journal of Abnormal Psychology, 97, 346-353.

47.

Wright

A. G. C.

Krueger

R. F.

Hobbs

M. J.

Markon

K. E.

Eaton

N. R.

Slade

(2013). The structure of psychopathology: Toward an expanded quantitative empirical model. Journal of Abnormal Psychology, 122, 281-294. doi:10.1037/a0030133

48.

Zvolensky

M. J.

Otto

M. W.

(2007). Special series affective intolerance, sensitivity, and processing: Advances in clinical science introduction. Behavior Therapy, 38, 228-233.