Temporal Sequencing of Change in Trauma-Related Beliefs and Therapeutic Alliance During Prolonged Exposure and Sertraline for Chronic PTSD

Abstract

Objective:

Changes in trauma-related beliefs (Cooper, Zoellner, et al., 2017) and therapeutic alliance (Crits-Christoph et al., 2011) have been found to temporally precede symptom reduction; however, it is likely these processes do not act in isolation but rather in interactive ways.

Methods:

The present study examined the temporal relationships between negative posttraumatic cognitions (PTCI; Foa et al., 1999) and therapeutic alliance (WAI; Tracey & Kokotovic, 1989) in 142 patients who were part of a randomized trial comparing prolonged exposure (PE) to sertraline for chronic PTSD.

Results:

Using time-lagged mixed regression models, improvements in the therapeutic alliance predicted subsequent improvements in trauma-related beliefs (d = 0.59), an effect accounted for by between-patient variability (d = 0.64) compared to within-patient variability (d = .04) giving weaker support to the causal role of alliance on outcome. Belief change did not predict improvements in alliance and neither model was moderated by treatment type.

Conclusion:

Findings suggest alliance may not be an independent driver of cognition change and point to the need for additional study of the impact of patient characteristics on treatment processes.

Posttraumatic stress disorder (PTSD) is a disabling condition that develops in a significant minority of individuals following exposure to a potentially traumatic event. Efficacious psychological interventions exist, including prolonged exposure therapy (PE; Foa et al., 2007), however, a substantial minority of individuals experience an incomplete or non-response to treatment (Bradley et al., 2005; Steenkamp et al., 2015). Although rates of dropout and nonresponse for PTSD treatment are on par with rates for other anxiety and depression disorders (Hans & Hiller, 2013; Hofmann & Suvak, 2006; Keijsers et al., 2001), a more nuanced understanding of why treatments work would enable clinicians to better optimize the therapeutic elements of treatment that lead to change and better understand the factors that lead to dropout or nonresponse (Kazdin, 2007; Murphy et al., 2009). Thus, understanding putative underlying change processes is critical for optimizing patient outcomes and reducing the burden of PTSD.

Negative trauma-related beliefs about oneself, others, and the world are a peritraumatic risk factor for the development of PTSD (Lancaster et al., 2011; Shahar et al., 2013). Negative posttraumatic cognitions diminish alongside PTSD symptoms (Foa & Rauch, 2004; Hagenaars et al., 2010; Nacasch et al., 2015) and have been subsequently recognized as a potential mediating variable related to change processes in PTSD treatment. Recent research using advanced statistical methods that consider time sequencing (e.g., lagged models) suggest negative posttraumatic belief change precedes PTSD symptom change (Cooper, Zoellner, et al., 2017; Kleim et al., 2013; Kumpula et al., 2017). These studies provide robust evidence for negative belief change as a mediator of PE, suggesting changes in how one thinks about oneself, others, and the world may be critical for reducing PTSD symptoms.

In considering the relationship between negative posttraumatic cognitions and PTSD symptoms, it is likely belief change interacts with or is facilitated by other putative mediators. The therapeutic alliance constitutes the collaborative nature of the therapeutic relationship between the psychotherapist and the patient throughout treatment (Bordin, 1979; Bordin, 1994). A stronger therapeutic alliance is associated with better treatment outcomes (e.g., symptom reduction, quality of life improvements) and greater therapeutic change across psychotherapies (Fluckiger et al., 2018; Horvath et al., 2011), including exposure-based therapies (Weck et al., 2016). Of note, only three studies to date directly investigated the temporal pattern of the therapeutic alliance in PE for PTSD. In a study examining patterns of alliance rupture-repair episodes among 116 patients treated with PE, unrepaired ruptures significantly predicted worse treatment outcomes (McLaughlin et al., 2014). In a sample of 61 adolescent girls, alliance improved substantially more over the course of PE than client-centered therapy (Capaldi et al., 2016); and in a sample of 65 adults randomized to standard PE or modified PE with imagery rescripting, Hoffart and colleagues (2013) found a slight within-person relationship between alliance component regarding the tasks of therapy and subsequent PTSD symptom change across both therapies. However, no research to date has examined the interactive relationship between the therapeutic alliance and belief change. It is quite plausible that the therapeutic alliance facilitates belief change by providing an opportunity for collaborative empiricism (Overholser, 2011; Tee & Kazantzis, 2011) and evidence for more realistic cognitions (e.g., “My therapist doesn’t think the rape was my fault and supports me- maybe there are some good, friendly people in the world.”). Alternatively, perhaps the formation, maintenance, and growth of therapeutic alliance is a result of belief change. A clear understanding of these processes and the associated temporal relationships between them will aid in better understanding how treatments work.

Randomized controlled trials offer a unique opportunity for the investigation of mediators, particularly when the treatment conditions are similarly effective, for it allows for direct comparisons across treatment condition (Kraemer et al., 2002; Laurenceau et al., 2007; Murphy et al., 2009). Sertraline, a selective serotonin reuptake inhibitors (SSRI), is one of only two medications approved by the U.S. Food and Drug Administration (FDA) for treatment of PTSD (Friedman & Davidson, 2014). Evidence suggests sertraline produces clinical global improvement (Brady et al., 2000; Davidson et al., 2001; Friedman et al., 2007; Lee et al., 2016; Zoellner et al., 2019; Zohar et al., 2002). Given the intracellular effects of SSRIs on neuronal deficits that contribute to psychopathology (Duman & Voleti, 2012), specifically the reduced prefrontal emotion regulatory brain regions observed in patients with depression (Anderson, 2000; Kanske et al., 2012) and PTSD (MacNamara et al., 2016; Stein et al., 2000), the therapeutic action of SSRIs may be substantially different from that of psychotherapy making it a suitable treatment comparison for the study of process variables in PE.

SSRIs also affect emotional and cognitive processes most often typified as psychological processes (Buhle et al., 2014; Harmer, 2008). In a study comparing antidepressant therapy to combined antidepressant and cognitive therapy for depression, few differences emerged in the change of self-reported core beliefs between treatments (Dozois et al., 2014). In contrast, another study found belief change predicted subsequent PTSD symptom improvement more strongly in PE (d = 0.93) compared to sertraline (d = 0.35), suggesting that change in beliefs may be a strong mediator in PE compared to SSRIs (Cooper, Zoellner, et al., 2017). However, no study to date has investigated the relationship between belief change and alliance in pharmacotherapy; and furthermore, no study to date has examined the therapeutic alliance in pharmacotherapy for PTSD. It is likely that alliance may serve as a more prominent mediator of change in PE than in sertraline given the greater role therapists serve in psychotherapy.

In this study, the temporal relationships between therapeutic alliance and belief change were examined in patients with PTSD treated with PE or sertraline using time lagged repeated-measures regression models to estimate the magnitude of the relationships between variables at the session level and to compare the direction of effect (e.g., whether alliance precedes cognitive change or vice versa running reverse models). We additionally disaggregated the raw scores for each variable into scores reflecting within-patient (e.g., alliance fluctuations occurring within a patient-therapist dyad during treatment) and between-patient variability (e.g., alliance fluctuations that reflect differences across dyads during treatment), allowing us to effectively control for stable between-patient differences by focusing on the potential relationship of within-patient processes of change and again, ran models in both directions to examine the direction of effect.

Although little study has focused on the relationship between alliance and belief change, change in negative cognitions consistently precede symptom change (Cooper, Zoellner, et al., 2017; Kleim et al., 2013; Kumpula et al., 2017; Zalta et al., 2014) and are a central emphasis in theoretical models underlying the development and treatment of PTSD (Ehlers & Clark, 2000; Foa et al., 2006). Given the literature underlying the notion that belief change precedes PTSD symptom change (e.g., Cooper, Zoellner, et al., 2017) and the mixed literature suggesting a possible effect of alliance on subsequent symptom change (e.g., Hoffart et al., 2013) we hypothesize that improvements in alliance will precede improvements in cognitions but not vice versa; an effect more prominent in PE than in sertraline.

Method

Participants

The sample for the current study (N = 142) was drawn from a large randomized controlled trial comparing PE with sertraline for the treatment of chronic PTSD (Zoellner et al., 2019). The trial consisted of two hundred men and women between the ages of 18 and 65 years with a primary diagnosis of chronic PTSD. Exclusion criteria were minimal to best maximize generalizability and included: primary DSM-IV diagnosis other than chronic PTSD; current diagnosis of schizophrenia or delusional disorder, medically unstable bipolar disorder, depression requiring immediate psychiatric treatment or with psychotic features; alcohol or substance dependence within the previous three months; severe self-injurious behavior or suicide attempt within the past three months; an ongoing relationship with perpetrator in cases of sexual or physical assault; medical contraindication for taking sertraline; or previous non-response to adequate trial of either PE (8 sessions or more) or sertraline (8 weeks, 150 mg/d). The trial was approved by Institutional Review Boards, and all participants provided written informed consent prior to participating.

For the present study, only patients who completed a minimum of four treatment sessions and who also had a minimum of three observations per measure were included (Yang & Maxwell, 2014), resulting in a final sample of 142 patients. See Table 1 for demographic characteristic breakdown by treatment type.

Table 1.

Demographic Characteristics in Full Sample and by Treatment Type

	Full Sample (n = 142)		PE (n = 86)		Sertraline (n = 56)
Characteristics	n	%	n	%	n	%
Female	111	78.2	71	82.6	40	71.4
Caucasian	100	70.4	59	68.6	41	73.2
Not college educated	97	68.3	53	61.6	44	78.6
Primary Trauma
Physical/sexual assault as adult	79	55.6	47	54.7	32	57.1
Physical/sexual assault as child	31	21.8	21	24.4	10	17.9
Accident or natural disaster	21	14.8	12	14.0	9	16.1
Combat	4	2.8	2	2.3	2	3.6
Death of loved one	7	4.9	4	4.7	3	5.4
	M	SD	M	SD	M	SD
Age	37.6	11.7	36.4	11.5	39.5	11.9
Time since target trauma	11.8	12.5	12.4	12.5	10.9	12.6

Note. PE = Prolonged Exposure; n = number; % = percent; M = mean; SD = standard deviation

Measures

PTSD Symptom Scale – Interview version (PSS-I) (Foa, Riggs, Dancu, & Rothbaum, 1993).

The PSS-I is a 17-item interviewer-rated measure of PTSD symptomatology used in this study to assess eligibility, diagnosis, and PTSD severity. The measure has demonstrated good reliability and convergent validity (Foa & Tolin, 2000). Diagnostic reliability of 10% of cases in this study was excellent (ICC = .99).

Structured Clinical interview for the DSM-IV (SCID-IV) (First, Spitzer, Gibbon, & Williams, 1995).

The SCID-IV was used to assess study eligibility and comorbid diagnoses. In this study, 10% of SCID-IVs were rerated for interrater reliability; reliability across diagnoses was acceptable (κ = .80).

Working Alliance Inventory-Patient- Short Form (WAI) (Tracey & Kokotovic, 1989).

The WAI evaluates the therapeutic alliance and comprises of 12 items scored on a 7-point Likert scale ranging from 1 (never) to 7 (always) with higher scores indicative of a stronger therapeutic alliance. Patients completed the WAI before sessions 2, 4, 6, 8, and 10 of treatment. The measure has demonstrated excellent reliability and internal consistency (Horvath & Greenberg, 1989). In this study, internal consistency at Session 2 (α = .93) and 2-week test-retest reliability (α = .81) was very good.

Posttraumatic Cognitions Inventory (Foa et al., 1999).

The PTCI is a 36-item self-report measure that assesses negative trauma-related beliefs about the self, the world, and self-blame. Items are scored on a 7-point Likert scale ranging from 1 (totally disagree) to 7 (totally agree), with higher scores indicative of greater negative cognitions. The measure has good test-retest reliability and excellent convergent validity, discriminant validity, and internal reliability (Foa et al., 1999). Patients completed the PTCI before every treatment session. In this study, internal consistency at pretreatment (α = .95) and 1-week test-retest reliability (α = .91) was excellent.

Procedure

Participants, recruited from community referrals and flyers, were initially screened via a semi-structured phone interview and scheduled for an intake evaluation upon determination of potential eligibility. After informed consent was obtained, an independent evaluator masked to treatment condition conducted an intake containing demographic and diagnostic information through structured interviews (PSS-I, SCID-IV) to determine potential eligibility. At the randomization visit, participants completed a battery of self-report measures including the PTCI. Patients received up to 10 weekly sessions of PE or sertraline. At post-treatment, an independent evaluator masked to treatment condition conducted a post-treatment evaluation including PTSD symptoms (PSS-I).

Treatment consisted of 10 weeks of PE provided by master’s or doctoral level clinicians, or sertraline provided by board certified psychiatrists. PE clinicians received a standardized clinical training prior to beginning the study, followed a standard treatment manual (Foa et al., 2002), and received weekly clinical supervision and consultation throughout the duration of the study. Sertraline was provided by board-certified psychiatrists for 10 weekly sessions of up to 30 minutes in duration, with up to 45 minutes for the first session. Patients were started on a dosage of 25 mg/day and were increased up to 200mg/day, if indicated and tolerated using a standard titration algorithm and manual (Marshall et al., 2001). The final mean dosage for this sample was 155 mg/day (SD = 59.0). Psychiatrists monitored side effects, adjusted medication dosage as needed, and provided general support but did not provide psychotherapy. Neither exposure or anti-exposure instructions were given.

Treatment sessions were audio recorded or videotaped and standard fidelity checklists were used by outside raters to assess protocol violations and adherence to required treatment components. Raters evaluated 10% of tapes and found no protocol violations for PE or sertraline. PE therapists completed 90% of essential components and psychiatrists completed 96%. PE sessions were also rated for therapist competence on a 3-point Likert scale ranging from 1 (inadequate) to 3 (adequate or better). Overall, PE therapist competence was very good (M = 2.73, SD = 0.32).

Data Analytic Plan

The sample was selected as a process research sample, seeking to uncover why and how PE and sertraline promote therapeutic change. Including patients who dropped out early in treatment and imputing clinical information for those patients, at best, provides little useful information (DeRubeis et al., 2014) and risks diluting or shadowing the effect of the mechanism in question or causing assumption violations such as with non-random missingness (Yang & Maxwell, 2014). In line with previous treatment process research (Kleim et al., 2013) and theoretical models that argue mixed-effect models in randomized longitudinal trials can produce biased estimates when missingness is missing not at random (Yang & Maxwell, 2014), only patients who completed a minimum of four treatment sessions were included. Additionally, and as further elaborated on below, a minimum of three observations per patient were needed for the disaggregation analyses resulting in a final sample of 142 patients.

To examine the relationship between alliance and belief change, time-lagged, repeated-measures regressions were utilized (Rovine & Walls, 2006), examining the directionality of the effect with the reverse model (e.g., whether alliance precedes cognitive change or vice versa). This statistical approach allows for an examination of potential temporal relationships between two variables, examining the strength of the relationship between a predictor at Time X, and a dependent variable at Time X + 1, while also controlling for the autocorrelation with that predictor at Time X, in line with theory (Curran & Bauer, 2011) and past method approaches (Cooper, Zoellner, et al., 2017; Strunk et al., 2012). Scores from the WAI (alliance) and PTCI (negative beliefs) included up to 5 time points: session 2 (Time 1), session 4 (Time 2), session 6 (Time 3), session 8 (Time 4), and session 10 (Time 5). For each variable, a set of dependent variables (Time 2 to Time 5) and a set of lagged predictors (Time 1 to Time 4) were combined in a single dataset. All models were tested in PROC Mixed with maximum likelihood in SAS 9.4 to account for missing variables. Two sets of analyses including time as a covariate were conducted: (1) WAI predicting PTCI and (2) PTCI predicting WAI.

Given the hierarchical nature of the treatment data, we sought to then disentangle within-subject effects from between-subject effects (Hoffman & Stawski, 2009). Between-patient effects refer to interindividual processes wherein change is investigated between groups of individuals; whereas, within-patient effects refer to intraindividual processes wherein change is investigated within individual patients. Within-patient variability identifies a relationship that cannot be attributed to stable patient characteristics (the between-patient variability) and thus is likely reflective of a potential causal process (Curran & Bauer, 2011). Research has largely drawn conclusions about within-patient processes from the analysis of between-patient data; however, the presence of a between-patient relationship is not adequate and sufficient evidence to claim a within-patient effect of a mediator on outcome (Kazdin, 2007). An observed effect could be a proxy for some other patient/subject level (between-patient) variable such as personality or diagnosis.

Thus, after running the initial time-lagged repeated measure regressions (termed the “aggregate” model), within-patient and between-patient variation for each measure were decomposed using procedures recommended by Curran and Bauer (2011). This process-oriented analytic approach has been commonly used (e.g., Braun et al., 2015; Sasso et al., 2016). For each hypothesized mediator—WAI and PTCI—we conducted a series of separate ordinary least squares (OLS) regressions for each patient in which we regressed each patient’s raw variable score on time (mean centered). We retained the time-specific residuals from each patient’s model (representing the deviation in each time-specific process score from the model-implied values) to obtain the within-patient score at each time point, and the patient-specific intercept from each patient’s model to obtain the between-patient score. As previously noted, this method requires a minimum of three observations per patient for a non-saturated model so that the number of data points exceeds the number of parameters being estimated. To illustrate using the WAI where t = time and i = a given patient, this would be represented as:

$$WA{I}_{ti}=b_{0i}+b_{1t}Tim{e}_{ti}+e_{ti}$$ (1)

$WA{I}_{ti}$ is the time-specific patient score (i.e., the patient’s alliance score at a given session), b_0i represents the model intercept (between-patient score), b_1t is the slope of the WAI scores across time, $Tim{e}_{ti}$ is the measure of time, and e_ti is the session-specific residuals from the model (within-patient score).

Next, we examined the within- and between-patient scores of the process variable (e_tiand b_0t from equation 1) simultaneously as predictors of change at the following time point. To illustrate, we entered the repeated measures of WAI-Within and WAI-Between (Time 1–4) as predictors of PTCI ($PTC{I}_{t+1i}$; Time 2–5) with PTCI at the current session ($PTC{I}_{ti}$) entered as a covariate as follows:

$$PTC{I}_{t+1i}={\beta }_0+{\beta }_1\left(PTC{I}_{ti}\right)+{\beta }_2\left(WA{I}_{Between}\right)+{\beta }_3\left(WA{I}_{Within}\right)+{\epsilon }_{ti}$$ (2)

The symbol β₀ represents the model intercept and ε_ti denotes the model error term. We specified time as the repeated variable and patient as the subject. Variance components was determined to have the best fit across all information criteria indices amongst covariance structures examined (viz. unstructured, toepliz, compound symmetry, autoregressive, and variance components). We specified maximum likelihood as the estimation method for covariance parameters (Allison, 2012) and between-within as the method for computing the denominator degrees of freedom. Signs were adjusted to facilitate interpretation such that a positive relationship always indicates more positive outcomes (i.e., lower symptom scores as indicated by the PTCI and higher ratings of alliance on the WAI). We additionally examined treatment (PE or sertraline) effects; if an interaction term was significant, we then ran the model separately within each treatment condition.

Results

Mean and standard deviation of patients’ process variable scores at each session can be seen in Table 2. Means and standard deviations for the within-patient and between-patient scores for each variable are provided in Table 3. Each patient only had one between-patient score (intercept from the model) per variable. For within-patient scores, each patient had five scores for each variable. Because residuals are parameterized to sum to zero (i.e., they represent the deviations between the model and the actual data), the mean for all within-patient scores was zero.

Table 2.

Means and Standard Deviations of Patients’ Observed Raw Alliance and Negative Posttraumatic Cognition Scores at Each Session for Patients in Both Conditions Combined and in Each Condition

Full Sample (n = 142)
Variable	Range	Session 2	Session 4	Session 6	Session 8	Session 10
WAI	12 – 84	65.44 (12.35)	67.01 (12.00)	68.97 (12.19)	70.73 (11.15)	71.02 (12.07)
PTCI	36 – 252	138.72 (40.00)	131.19 (43.17)	118.23 (44.05)	107.29 (42.13)	98.00 (40.83)
Prolonged Exposure (n = 86)
WAI	12 – 84	67.40 (10.25)	69.24 (10.30)	70.92 (10.77)	70.40 (11.62)	72.64 (9.73)
PTCI	36 – 252	139.66 (38.85)	132.99 (43.28)	119.30 (42.39)	106.22 (41.36)	91.92 (35.76)
Sertraline (n = 56)
WAI	12 – 84	62.04 (14.85)	63.29 (13.81)	65.13 (13.95)	71.38 (10.27)	68.81 (14.49)
PTCI	36 – 252	137.07 (42.32)	127.93 (43.30)	116.03 (47.77)	109.38 (44.08)	106.38 (45.99)

Note. Means are reported outside of the parentheses, standard deviations reported in parentheses WAI = Working Alliance Inventory; PTCI = Posttraumatic Cognitions Inventory.

Table 3.

Means and Standard Deviations of the Within- and Between-Patient Scores in Both Conditions Combined and in Each Condition

	PE & Sertraline (n = 142)		PE (n = 86)		Sertraline (n = 56)
	Between	Within	Between	Within	Between	Within
WAI	68.88 (11.72)	0 (3.56)	70.90 (10.16)	0 (3.16)	65.83 (13.19)	0 (4.17)
PTCI	116.55 (37.98)	0 (11.36)	116.66 (36.45)	0 (11.99)	116.38 (40.23)	0 (10.17)

Note. Means are reported outside of the parentheses, standard deviations reported in parentheses. WAI = Working Alliance Inventory; PTCI = Posttraumatic Cognitions Inventory. Scores were derived utilizing a residualized group-mean centering procedure for detrended estimates of within-patient variation. Because residuals are parameterized to sum to zero, the within-patient scores each have a mean of 0. For the SD of within-patient scores, we calculated the SD for each patient and averaged those SDs.

Aggregated Results

For the model with belief change as the dependent variable, the cross-lagged effect of WAI on PTCI was statistically significant (d = 0.59), suggesting that improvements in alliance generally preceded improvements in cognitions in the aggregated model. For the model with alliance as the dependent variable, the cross-lagged effect of PTCI change in predicting subsequent WAI improvements was not significant and negligible (d = 0.07).

Disaggregated Results

As seen in Table 4 showing the effects of within- and between-patient variability in scores, the majority of the variance in the model predicting PTCI from time-lagged WAI could be explained by between-patient WAI scores (d = 0.64); whereas, within-patient WAI scores did not significantly predict subsequent PTCI scores (d = 0.04). In the reverse model, neither within-patient or between-patient PTCI scores significantly predicted WAI scores.

Table 4.

Time Lagged Multilevel Regressions of Alliance (Working Alliance Inventory; WAI) and Beliefs (Post-Traumatic Cognitions Inventory; PTCI) in the Aggregated Model and Disaggregated Model (Within and Between Effects)

Variable	b	SE	t	p	d
Predicting PTCI from time-lagged WAI
Aggregated WAI	0.43	0.09	4.93	<.0001	0.59
Within WAI	0.08	0.26	0.30	0.77	0.04
Between WAI	0.49	0.09	5.39	<.0001	0.64
Predicting WAI from time-lagged PTCI
Aggregated PTCI	0.01	0.01	0.61	0.54	0.07
Within PTCI	0.01	0.03	0.41	0.68	0.05
Between PTCI	0.02	0.01	1.76	0.08	0.21

Note. WAI = Working Alliance Inventory; PTCI = Posttraumatic Cognitions Inventory; d = Cohen’s d, where d = t*√(2/n); N = 142

Treatment Effects

To examine whether the relationship between alliance and belief change differed by treatment modality, we tested treatment X within- and between-predictor interactions in each model. In the model predicting change in PTCI, the treatment X within-patient WAI and the treatment X between-patient WAI interactions were not significant (ps = .49 and .32 respectively). Similarly, in the model predicting change in WAI, neither the Treatment X within-patient PTCI interaction (p =.46) or the treatment X between-patient PTCI (p = .59) were significant, indicating that there were no significant differences between treatments in the relationship between these constructs.¹

Discussion

Prior to disentangling within- and between-patient effects, therapeutic alliance predicted improvements in negative posttraumatic cognitions but not vice versa, suggesting alliance may be an independent driver of cognition change; however this was not the case. While we are unaware of any previous study that directly examined the temporal relationship between the alliance and cognitions, previous studies investigating the temporal relationship between alliance and symptom change have been mixed with some studies showing alliance precedes symptom improvements (e.g., Crits-Christoph et al., 2011; Held et al., 2022; Xu & Tracey, 2015; Zilcha-Mano et al., 2014) whereas others did not find support for this relationship (e.g., Hillman et al., 2022; Renner et al., 2018; Santoft et al., 2018; Strunk et al., 2012).

Notably, the effect of alliance on cognitive change in the current study was entirely accounted for by the between-patient variability in alliance and not moderated by treatment. Because within-patient estimates are unlikely to be confounded with stable person characteristics, within-patient effects are more likely to reflect causal processes than between-patient effects. The study findings thus suggest that the identified relationship observed in the aggregated model is less likely to be reflective of a potential mechanistic process. While the alliance-outcome correlation has consistently been demonstrated (Fluckiger et al., 2018; Horvath et al., 2011), less research has investigated the temporal within-patient relationship between alliance and outcome. In a recent systematic review examining the alliance-outcome relationship for individual psychotherapy across 37 studies, Baier and colleagues (2020) found that only 6 studies (Baldwin et al., 2007; Falkenström et al., 2016; Falkenström et al., 2013; Gómez Penedo et al., 2020; Rubel et al., 2019; Sasso et al., 2016) adequately disaggregated within and between-patient effects with mixed findings and significant heterogeneity between study designs, patient population, psychotherapy modality, and statistical analytic procedures thereby limiting conclusions.

In the present data, alliance does not appear to be a prominent mediator of cognition change and thus may not be a primary mechanism for facilitating belief change. Additionally, because there was no significant relationship between the within-patient cognition scores and subsequent alliance, the results suggest that a lack of reduction in cognitions may not be indicative of a poor alliance. The practice of developing a strong alliance is largely argued as a “nonspecific” and “transtheoretical” factor –that is, a necessary precondition for any successful therapy (Hatcher & Barends, 2006; Raykos et al., 2014; Weck et al., 2015). As an important component of any successfully psychotherapy or pharmacotherapy, the treatment providers in this study were likely quite attuned to the therapeutic relationship; alliance ratings were high across sessions with limited variability across sessions (WAI: 65.44 – 71.02; see Table 2). It is also plausible that the alliance may play a causal role in outcome by facilitating other processes—such as hope, treatment expectancy, and buy-in—that may then help to drive cognition change as opposed to a direct relationship between alliance and cognition change. Indeed, the correlation between alliance and expectancy of treatment outcome assessed at session 1 in the data was r = 0.55 (p <.001). Thus, while alliance may not be a driver of cognition change in this study, the hypothesized transtheoretical nature of the alliance, coupled with the finding that results were not moderated by treatment, suggests that it nevertheless should be attended to by psychotherapists and prescribing providers alike.

As the results show, attending to or not attending to the between-patient and within-patient variability in scores can alter findings and have a substantial impact on conclusions drawn from data. After disaggregating the variables, improvements in within-patient alliance did not significantly predict next session negative belief change. Importantly, disparate findings were found in the aggregate model providing a tangible illustration as to how effects can be masked making a Type II error or, falsely identified making a Type I error when attention is not directed to parsing out the variation likely confounded by stable patient traits (between-patient effect) versus the variation that is more likely to reflect a true causal relationship (within-patient effect).

Given the strength in variability of between-patient effects, additional research will need to examine the extent to which patient characteristics play a role in treatment outcome and perhaps moderate various putative mechanisms of change. It could be that patient characteristics have a differential effect on different mechanisms such that treatments and specific techniques (e.g., in vivo homework, imaginal exposure) might be able to be matched with certain kinds of patients and augmented to meet the needs of individual patients. In other words, while treatments might work on a diverse group of patients, they may not work in the same way for every patient (Cooper, Clifton, et al., 2017). Furthermore, it is possible the between-patient alliance mediates person-level characteristics rather than being confounding with such stable patient traits which would not invalidate the possible causal role of alliance. By understanding the specifics underlying change, clinicians ought to be able to tailor treatments to meet the needs of individual patients thus personalizing care, and theoretically, advancing the success of such interventions.

Key strengths of this study include a robust analytic strategy that disentangled between-patient and within-patient effects, stringent study sample selection criteria that ensured patients included achieved an adequate “dose” of treatment, and a large and diverse sample of male and female patients with heterogeneous trauma exposure. However, a number of limitations should be noted. Due to the timing of the data collection with various measures, analyses utilized data from sessions 2, 4, 6, 8, and 10 and thus we did not examine change after each session but rather change after every two sessions. However, with five data points, the models remained well-suited for examining disaggregated change over time. Furthermore, though linear models fit the data, it is also possible that change in alliance may be more complicated. As noted by others (Sasso et al., 2016), limited within-patient variability in raw variable scores could reduce power to detect within-patient process-outcome relations and thus reduce the strength of relationships. An examination of the mean alliance range over time (WAI: 65.44 – 71.02; see Table 2) revealed limited variability which could be attributed to a true lack of within-patient variability, or, insensitivity of the measure to capture within-patient variability in alliance. Finally, a simulation study by Falkenström et al. (2022) revealed severely biased estimates of cross-lagged effects for multilevel models (MLM) and thus recommended the use of disaggregated structural equation models (SEM). Accordingly, we confirmed the stability of the present MLM results by reanalyzing the data using latent curve modeling with structured residuals (Curran et al., 2014). There were no substantive differences in results between the models. Nevertheless, the challenges of modeling dynamic relationships between two constructs over time is well documented (Curran & Hancock, 2021). Different models may reveal drastically different findings highlighting the critical importance of theory in model selection as well replication of findings both across models and in future research.

In sum, the present findings imply that how connected patients feel with their providers may not be an independent driver of cognition change specifically. Future research is needed to aid in understanding the potential mediating role of alliance on symptom change and other hypothesized treatment processes. Additionally, given the strength in variability of the between-patient alliance score on belief change, additional research should examine potential moderators of the alliance-outcome relationship such as personality and comorbid diagnoses. Perhaps mechanisms of change in PTSD treatment are more patient-specific requiring a greater nuanced understanding on both intra-individual and inter-individual change processes. By further studying the role of alliance in treatment outcome, therapists will be able to best optimize and personalize treatments to specific patients.

Clinical or Methodological Significance of this Article.

Studying relationships involving between-patient and within-patient process variables are important for understanding how treatments work. In this study, the temporal relationship between the therapeutic alliance and cognition change was entirely accounted for by between-patient variability suggesting the relationship observed may be better attributed to confounding stable patient characteristics rather than a potential causal process. This emphasizes the methodological importance of disaggregating between and within-patient effects and also suggests that therapeutic alliance may not be an independent driver of cognition change.