Prolonged exposure in routine outpatient care: a naturalistic study of feasibility, safety and effectiveness

ABSTRACT

Background: Prolonged Exposure (PE) is a first-line evidence-based treatment for Posttraumatic Stress Disorder (PTSD), with well-established efficacy in randomised controlled trials. However, its implementation in routine care remains limited, particularly in outpatient clinics where evidence-based psychotherapies are not widely adopted.

Objective: This study aimed to evaluate the feasibility, safety and effectiveness of PE therapy delivered in a real-world outpatient setting.

Method: Data were obtained from 95 adults diagnosed with PTSD using the Clinician-Administered PTSD Scale for DSM-5 (CAPS-5) who received PE in a French outpatient clinic. The sample was characterised by a mean age of 35.65 years (SD = 13.16), 76% women (n = 72), a mean CAPS-5 score of 38 (SD = 9.4), and 61% reporting sexual trauma as the index event. PTSD Checklist for DSM-5 (PCL-5) and Quick Inventory of Depressive Symptomatology – Self-Report (QIDS-16-SR) were assessed at intake, pre-treatment, post-treatment and three-month follow-up. Mean changes were estimated using a linear mixed-effects model.

Results: Patients received an average of 8 sessions of PE (SD = 3.75). Scores consistent with probable remission (PCL-5 < 33) was observed in 75.53% of patients, and 79.78% demonstrated a clinically significant improvement (≥10-point reduction on the PCL-5). A large decrease in PTSD symptoms was observed from pre- to post-treatment (Δ = 25.57, d = 1.88, p < .001), and to 3-month follow-up (d = 2.17, p < .001), exceeding the small effect found during the waiting-list phase (d = 0.33, p = .01). Similar improvements were found for depressive symptoms. Only 1% of patients showed a reliable clinical symptom worsening (≥10-points increase on the PCL-5) and the dropout rate was 9.47% (fewer than eight sessions completed without remission).

Conclusion: These findings suggest that PE may be feasibly and safely delivered in routine outpatient care and may be associated with substantial reductions in PTSD symptoms. They highlight the potential for successful Implementation of evidence-based trauma treatments in real-world clinical settings where structured psychotherapies are not yet widely integrated.

HIGHLIGHTS

• Prolonged Exposure may be implemented in routine setting with relatively low usage of evidence-based treatments.

• Results indicate outcomes consistent with those observed in meta-analysis.

• Prolonged Exposure may be safely proposed within routine outpatient clinical practice.

1. Introduction

Posttraumatic Stress Disorder (PTSD) is a prevalent and disabling condition. Lifetime prevalence in the general population is estimated at approximately 3.9% worldwide (Koenen et al., 2017), and prevalence following trauma exposure may reach 23.95% (Schincariol et al., 2024). PTSD is associated with substantial clinical burden, including elevated risk of suicidal ideation, suicide attempts, and death by suicide (Akbar et al., 2023), as well as significant economic costs to healthcare systems and society (Davis et al., 2022).

Since the 1990s, several psychotherapies for PTSD have been developed, some of which are now recommended as first-line treatments (Charney et al., 2018). Among those, Prolonged Exposure (PE) (Foa et al., 2019) is a manualized trauma-focused therapy based on Emotional processing theory (Foa & Kozak, 1986) typically delivered in 8–15 weekly 90-minute sessions comprising psychoeducation, in-vivo exposure and imaginal exposure (McLean & Foa, 2024). Meta-analyses consistently support the efficacy of PE with large effect sizes (Cusack et al., 2016; McLean et al., 2022; Powers et al., 2010) and an average remission rate of approximately 66% (Milligan et al., 2025). PE has also demonstrated effectiveness in patients with comorbid conditions such as depression, anxiety or substance use disorders (Hoeboer et al., 2024; van Minnen et al., 2015). However, approximately one in five patients discontinue treatment before its completion (Varker et al., 2021).

Despite strong evidence supporting PE and its recommendation as a first-line treatment across major international clinical guidelines (APA, 2025; ISTSS, 2018; NICE, 2018; Phoenix Australia, 2021; VA/DoD, 2023), PE remains underutilised in routine clinical practice. many clinicians remain reluctant to use exposure-based interventions for PTSD and implementation barriers include clinician beliefs regarding exposure therapy and limited access to training (Becker et al., 2004; McLean & Foa, 2013; Racz et al., 2024; van Minnen et al., 2010). These barriers may be particularly relevant in France, where PE is not explicitly recommended in national guidelines (Haute Autorité de Santé, 2007) and no structured national PE training programmes are currently available. To our knowledge, no French study has examined the implementation of evidence-based trauma-focused treatments in routine care; consequently, little is known about their dissemination, accessibility, or clinical outcomes within the French mental health system. Indirect evidence suggests that psychodynamic and non-manualized approaches continue to occupy an important place in French psychiatric training and practice (Botbol & Gourbil, 2018), whereas behavioural and cognitive-behavioural therapies have historically faced cultural and institutional skepticism (Amouroux, 2017), which may contribute to negative beliefs regarding exposure therapy.

Since 2021, the Normandy Center for Psychotrauma (Centre Régional Psychotraumatisme de Normandie; CRPN) has implemented PE in routine clinical practice as part of one of the first structured efforts to integrate trauma-focused evidence-based psychotherapy into routine care in France. This naturalistic context provides an opportunity to examine the feasibility, safety, and effectiveness of PE in routine outpatient care. To address this question, we conducted a pragmatic retrospective longitudinal study based on routinely collected clinical data from the CRPN treatment programme. Findings may help inform future dissemination and implementation efforts for trauma-focused evidence-based treatments in healthcare systems where such approaches remain underutilised.

Based on prior clinical trials and meta-analyses, we hypothesised a substantial reduction in PTSD symptom severity from baseline to endpoint, with a large effect size (Powers et al., 2010). We further expected symptom reduction to be greater during the treatment phase than during the waiting period. Consistent with previous findings, we also hypothesised maintenance of treatment gains at follow-up, characterised by a large effect size between baseline and follow-up assessments (Kline et al., 2018). In addition, we expected that 55%–75% of participants would score below the clinical cutoff on the PCL-5 following treatment (Milligan et al., 2025).

We also hypothesised medium-to-large reductions in depressive symptoms, with maintenance of gains at follow-up (van Minnen et al., 2015). Regarding feasibility and safety, we expected a dropout rate ≤20% (Varker et al., 2021), with fewer than 5% of participants demonstrating clinically significant symptom worsening.

Finally, exploratory analyses were conducted to examine PTSD symptom trajectories across treatment sessions, particularly around sessions 8 and 12, corresponding to typical PE treatment durations (NICE, 2018).

2. Method

2.1. Research setting

This study took place at the CRPN between 2021 and 2025. The local research ethics committee of the University Hospital of Caen Normandy reviewed and approved this repository-based protocol (#3109), which authorised the retrospective use of routinely collected clinical data from adult patients assessed or treated for PTSD at the CRPN. In accordance with French regulations governing non-interventional research based on routinely collected healthcare data, specific written informed consent for research participation was not required. All procedures were conducted in accordance with institutional ethical standards and the Declaration of Helsinki. Data were retrospectively analyzed using anonymized clinical records. The CRPN is a specialised outpatient trauma clinic embedded within Caen University Hospital and integrated into the French public healthcare system. It is one of several regional psychotrauma centres established nationally to improve access to specialised trauma-related care (El-Hage et al., 2019). Patients may directly self-refer, and treatment is publicly funded and provided at no direct cost.

2.2. Participants

For this study, we retrospectively reviewed available routine-care clinical records and included the first 100 accessible patient files meeting eligibility criteria for PE at the CRPN. Requirements to receive PE at the CRPN required a DSM-5 PTSD diagnosis established with the Clinician-Administered PTSD Scale for DSM-5 (CAPS-5; Weathers et al., 2018), as well as a CAPS-5 total severity score > 20. This severity threshold reflects the clinical organisation of the CRPN as a specialised outpatient trauma centre operating under limited treatment capacity, where priority is generally given to individuals presenting with at least moderate PTSD symptom severity. Patients were not eligible to receive PE if another current condition required immediate or prioritised treatment, such as an acute or unstable psychotic or bipolar disorder, substance-use disorder requiring medical stabilisation, or clinically significant suicide risk requiring a higher level of care. Eligibility to PE was assessed during the initial clinical evaluation conducted by trained clinicians (psychiatrists or clinical psychologists). Criteria of non-eligibility were evaluated through a clinical assessment, and these decisions were routinely discussed within the clinical team during weekly case consultations. No other comorbid psychiatric conditions were considered exclusionary, consistent with real-world clinical practice. Additional exclusion criteria for the present study included having received another first-line trauma-focused therapy at the CRPN (e.g. Cognitive Processing Therapy) and being under 18 years of age. Among the first 100 accessible patient files, five met exclusion criteria. Three patients were excluded because they were under 18 years of age, and two were excluded because they had received an alternative first-line trauma-focused therapy at the centre. The final sample comprised 95 participants with a mean age of 35.84 years (SD = 13.37; range: 18–68 years; median = 33.07), and the majority were female (76%). The mostly reported index trauma was sexual assault (60%) followed by physical assault (16.84%) with a median time since trauma of 6.16 years (min = 0.58; max = 63). Additional Clinical and socio demographic characteristics, as well as index trauma types based on the Life Events Checklist for DSM-5 (LEC-5; Weathers et al., 2013), are detailed in Table 1.

Table 1.

Demographic and clinical characteristics of a sample of individuals with PTSD receiving at least one session of prolonged exposure (N = 95).

Variable	n	Statistic
Age (years), M (SD)	95	35.84 (13.37)
Gender, % (n)	95
Male		24.21% (23)
Female		75.79% (72)
Index Trauma, % (n)	95
Sexual assault		60% (57)
Physical assault		16.84% (16)
Transport accident		9.47% (9)
Violent death		4.2% (4)
Assault with a weapon		3.2% (3)
Severe human suffering		2.11% (2)
Injury or illness		1.05% (1)
Fire-Explosion		1.05% (1)
Sudden death		1.05% (1)
Serious injury, harm, or death caused to someone else		1.05% (1)
Time since trauma (years), Mdn (min; max)	81	6.12 (0.58; 63)
Baseline CAPS-5 total score, M (SD)	95	38.54 (9.53)
Baseline QIDS-16-SR total score, M (SD)	95	14.76 (5.36)

Note: M = mean; SD = standard deviation; Mdn = median; CAPS-5 = Clinician-Administered PTSD Scale for DSM-5; QIDS-16-SR = Quick Inventory of Depressive Symptomatology Self-Report. The ‘sexual assault’ category included rape, attempted rape, forced sexual acts, and other unwanted sexual experiences.

2.3. Procedures

This pragmatic retrospective longitudinal study was based on routinely collected clinical data from a real-world outpatient trauma clinic. Patients were referred by healthcare professionals or self-referred and underwent a standardised intake evaluation to determine eligibility for trauma-focused therapy. Those meeting inclusion criteria were placed on a waiting list until a therapist became available, reflecting the natural organisation of care in the clinic.

PE was delivered individually, in person, on a weekly basis, in 90-minute sessions by psychiatrists (n = 2) and Master’s-level clinical psychologists (n = 8). Therapists had limited prior clinical experience overall, with a median of 0.5 years of post-graduate clinical experience (min = 0; max = 22; mean = 4.4, SD = 8.6), reflecting a highly skewed distribution driven by two highly experienced clinicians. None had prior experience delivering PE before the study. Five clinicians had prior formal training in cognitive-behavioural therapy, whereas four had integrative or psychodynamic orientations and one had training in neuropsychology. Four therapists completed the official 3-day online PE training and subsequently trained the remaining therapists internally. All therapists followed the PE treatment manual (Foa et al., 2019) throughout the study. In addition, therapists participated in weekly 1-hour in-person peer case consultation meetings to discuss clinical challenges and support treatment delivery. Participant flow and assessment availability across study time points are presented in Figure 1. At intake (T0), patients were assessed with the CAPS-5 (Weathers et al., 2018), the Life Events Checklist for DSM-5 (LEC-5; Weathers et al., 2013), the PTSD Checklist for DSM-5 (PCL-5; Blevins et al., 2015), as well as the Quick Inventory of Depressive Symptomatology (QIDS-16-SR; Reilly et al., 2015). The PCL-5 and QIDS-16-SR were again assessed at each session from the baseline (T1) (n = 95; first PE session) to the endpoint (T2) (n = 95; last PE session) and 3-month follow-up (T3) (n = 35). The mean duration of the waiting-list phase (from T0 to T1) was 120.71 days (n = 95, SD = 72.26; range: 13–420; median = 106), and the treatment phase (from T1 to T2) averaged 77.7 days (n = 90, SD = 41; range: 4–252; median = 76.5), with a mean of 8.11 sessions (n = 95, SD = 3.76; range: 1–18; median = 8). Follow-up visits (T3) were conducted when clinically feasible, on average 95.91 days (n = 35, SD = 17.11; range: 59–136; median = 92) after treatment completion.

Figure 1.

Participant flow through the study, including exclusions and available assessments at each time point.

2.4. Measures

The primary outcome was PTSD symptom severity, assessed with the PTSD Checklist for DSM-5 (PCL-5). Secondary outcomes included depressive symptom severity assessed with the Quick Inventory of Depressive Symptomatology Self-Report (QIDS-16-SR). Diagnostic assessment of PTSD and trauma exposure were conducted using the Clinician-Administered PTSD Scale for DSM-5 (CAPS-5) and the Life Events Checklist for DSM-5 (LEC-5), respectively. All self-report measures were administered in paper format as part of routine clinical care.

The French-validated version of the Clinician-Administered PTSD Scale for DSM-5 (CAPS-5; Weathers et al., 2018; French validation: Rivest-Beauregard et al., 2022) was used to establish PTSD diagnosis at intake. The CAPS-5 is a 20-item semi-structured clinician interview assessing DSM-5 PTSD symptoms over the past month. Items are rated on a 5-point severity scale ranging from 0 (‘absent’) to 4 (‘extreme/incapacitating’). The CAPS-5 is considered the gold standard for PTSD assessment and has demonstrated strong reliability and validity. A positive diagnosis required fulfilment of DSM-5 symptom criteria. Total severity scores range from 0 to 80, with higher scores indicating greater PTSD symptom severity.

Traumatic events were evaluated with the Life Events Checklist for DSM-5 (LEC-5; Weathers et al., 2013), which assesses 16 categories of potentially traumatic events and the mode of exposure (e.g. direct, witnessed, learned about). PTSD symptom severity was assessed using the PTSD Checklist for DSM-5 (PCL-5; Blevins et al., 2015), a 20-item self-report questionnaire assessing PTSD symptoms experienced over the past week. Items are rated on a 5-point Likert scale ranging from 0 (‘not at all’) to 4 (‘extremely’), yielding total scores ranging from 0 to 80, with higher scores indicating greater symptom severity. The PCL-5 has demonstrated excellent reliability and validity, including in its French-validated version (Ashbaugh et al., 2016; Forkus et al., 2023). The PCL-5 demonstrated good internal consistency in the current sample (Cronbach’s α = .84). Clinically reliable change was defined as a change of ≥10 points on the PCL-5 (Blanchard et al., 2023), with decreases indicating clinically reliable improvement and increases indicating reliable symptom exacerbation. Probable remission was defined as a PCL-5 score <33 (Forkus et al., 2023). Depressive symptoms were assessed using the Quick Inventory of Depressive Symptomatology – Self-Report (QIDS-16-SR; Reilly et al., 2015), a 16-item self-report questionnaire assessing depressive symptoms over the previous seven days. Items are rated on a 4-point scale ranging from 0 to 3, with total scores ranging from 0 to 27. Higher scores indicate greater depressive symptom severity. The QIDS-16-SR has demonstrated good psychometric properties, including satisfactory reliability and validity. However, to our knowledge, no formal French validation study of the QIDS-16-SR has been published to date.

2.5. Statistical analysis

To examine the effectiveness of PE, we relied on a within-subject delayed-treatment control, with the period between the intake (T0) and baseline (first PE session; T1) constituting a naturalistic waiting-list control phase during which no trauma-focused intervention was delivered, and the period between baseline (T1) and endpoint (last PE session; T2) representing the active treatment phase. The post-treatment endpoint was defined as the last observed assessment at each participant’s final PE session, reflecting the end of the treatment in this naturalistic setting where treatment duration varied across individuals.

PE effectiveness was examined using both statistical and clinical significance indices. Statistical indices included standardised effect sizes (Cohen’s d; Cohen, 1988) and t-tests from planned contrasts derived from estimated marginal means. These marginal means were estimated using linear mixed-effects models (LMMs) with time entered as a categorical repeated-measures factor. Separate models were fitted for PCL-5 and QIDS-16-SR scores. Both models included a random intercept to account for within-subject dependency. For all models, visual inspection of diagnostic plots indicated that the assumptions of residual normality, homoscedasticity, and normality of random effects deviations were reasonably met.

For both outcomes, we first compared a random-intercept-only model with a model including random slopes for time. In both cases, the random-slope model was not retained because it was overparameterized relative to the available data, as indicated by the number of random-effects parameters exceeding the number of observations. Because models including random slopes were not supported, alternative residual variance–covariance structures were examined to account for within-subject dependency.

For the PCL-5 model, we compared alternative residual variance–covariance structures, independence (AIC = 2469.00, BIC = 2491.46), compound symmetry (AIC = 2471.00, BIC = 2497.20), autoregressive AR(1) (AIC = 2468.49, BIC = 2494.70), and unstructured (AIC = 2463.11, BIC = 2519.25). Because alternative covariance structures did not provide meaningful improvement in model fit, and considering the principle of parsimony, the independence structure was retained.

For the QIDS-16-SR model, fit indices were as follows: independence (AIC = 1761.996, BIC = 1784.258), compound symmetry (AIC = 1763.996, BIC = 1789.969), autoregressive AR(1) (AIC = 1763.036, BIC = 1789.009), and unstructured (AIC = 1774.514, BIC = 1830.171). None of the alternative structures improved model fit relative to the independence model, as reflected by consistently higher AIC and BIC values. Therefore, the independence structure was retained as the most parsimonious model.

To model symptom trajectories over time, a series of nested models were compared using maximum likelihood estimation. First, a random-intercept-only model was compared with a model including random slopes for time, and the inclusion of random slopes substantially improved model fit (AIC = 5595.53 vs. 5803.05; BIC = 5623.22 vs. 5821.50; p < .001), indicating significant inter-individual variability in change over time. Second, a quadratic term for time was added to account for potential non-linear trajectories, which significantly improved model fit (AIC = 5785.91 vs. 5803.05; BIC = 5808.98 vs. 5821.50), suggesting that symptom change over time was not strictly linear. The final model included both linear and quadratic time effects with random slopes for time. Given this specification, within-subject correlation was assumed to be adequately captured by the random-effects structure, and no additional residual covariance structures were included in these trajectory models.

For all analysis, effect sizes were computed following Feingold (2009), defined as the model-based contrast divided by the baseline within-group standard deviation of the raw scores. Effect sizes (Cohen’s d) were interpreted as follows: <0.20 = negligible, 0.20–0.49 = small, 0.50–0.79 = medium, >0.80 = large (Cohen, 1988). Statistical significance was set at α = .05 (two-tailed) for all analyses.

Given the flexible and naturalistic nature of treatment delivery, dropout was defined as discontinuing PE before eight sessions while remaining symptomatic (PCL-5 ≥ 33 at the last available assessment). The eight-session threshold was chosen to approximate a typical course of PE delivered in routine care (NICE, 2018).

To examine the plausibility of the missing-at-random (MAR) assumption for follow-up PCL-5 data, we compared PCL-5 scores between participants with and without missing follow-up data at each time point, as well as the number of sessions, using linear regression models.

All analyses were conducted in R (version 4.5.0) within RStudio. LMMs were fitted using the lme4 package (Bates et al., 2015) and nlme package (Pinheiro & Bates, 2026). Estimated marginal means (EMMs) and planned contrasts were obtained using the emmeans package (Lenth & Piaskowski, 2026).

At the item level, missing responses on the PCL-5 were handled using a prorating procedure. When fewer than 20% of items were missing, total scores were computed by multiplying the mean of available items by the total number of items. If more than 20% of items were missing, the total score was considered missing. Across the full dataset, 8.28% of total scores were missing, compared to 0.75% after applying the prorating rule, indicating a low level of item-level missingness. The proportion of missing data varied slightly depending on the subset of observations included in each analysis.

3. Results

Descriptive statistics for the main study variables at each time point are presented in Table 2.

Table 2.

Descriptive statistics for main study variables at each time point.

Variable	Time point	n	Mean	SD
PCL-5	T0	95	53.23	11.97
PCL-5	T1	95	49.32	13.62
PCL-5	T2	94	23.76	16.22
PCL-5	T3	35	19.94	16.20
QIDS	T0	95	14.77	5.36
QIDS	T1	84	14.75	5.42
QIDS	T2	93	8.51	5.48
QIDS	T3	35	8.57	6.20

3.1. Effectiveness

The linear mixed-effects model indicated a significant effect of time on PTSD symptoms (F = 191.37, p < .001). The low intraclass correlation coefficient (ICC = 0.06) suggested that most variance reflected within-person change over time rather than stable between-patient differences. Figure 2 illustrates the modelled trajectory of PCL-5 scores across assessment points, and detailed estimated marginal means and planned contrasts are presented in Tables 3 and 4. PTSD symptoms showed a small but significant reduction during the waiting period (d = 0.32, p = .01), followed by a large reduction during treatment (d = 1.88, p < .001). The contrast comparing symptom change during the waiting period and treatment phase was significant (d = 1.59, p < .001), indicating substantially greater improvement during PE than during the waiting period. Treatment gains were maintained at 3-month follow-up, with a large reduction relative to baseline (d = 2.17, p < .001) and no significant symptom increase between post-treatment and follow-up assessments (d = 0.25, p = .07).

Figure 2.

Estimated means in PCL-5 and QIDS with 95% confidence intervals across time points.

Note: Left panel report estimated marginal means for PCL-5. Right panel shows estimated marginal means for QIDS-16-SR. Error bars represent 95% confidence intervals.

Table 3.

Estimated means for PCL-5 and QIDS-16-SR over the course of PE treatment.

Outcome	Time point	n	Mean	SE	df	95% CI
PCL-5	T0	95	53.2	1.46	207	[50.3–56.1]
PCL-5	T1	95	49.32	1.46	207	[46.4–52.2]
PCL-5	T2	94	23.7	1.47	208	[20.9–26.6]
PCL-5	T3	35	19.7	2.17	313	[15.4–24.0]
QIDS	T0	95	14.77	0.562	161	[13.66–15.88]
QIDS	T1	84	14.79	0.580	176	[13.65–15.94]
QIDS	T2	93	8.56	0.565	163	[7.44–9.67]
QIDS	T3	35	8.56	0.760	284	[7.06–10.06]

Degrees of freedom method: Kenward-Roger.

Table 4.

Contrasts and effect sizes for PCL-5 and QIDS-16-SR.

Outcome	Contrast	n	Estimate Δ	SE	df	t-ratio	p-value	Cohen’s d [95% CI]
PCL-5	T0–T1	95	3.91	1.51	220	2.59	.01	0.32 [0.07; 0.57]
PCL-5	T1–T2	94	25.58	1.51	220	16.883	<.001	1.88 [1.66; 2.09]
PCL-5	T2–T3	34	4.06	2.20	237	1.84	.07	0.25 [−0.02;0.51]
PCL-5	T1–T3	34	29.63	2.20	237	13.48	<.001	2.17 [1.86; 2.50]
PCL-5	Treat vs wait	94	21.67	2.62	220	8.27	<.001	1.59 [1.22;1.97]
QIDS	T0–T1	84	0.02	0.51	211	0.05	.96	0.01 [−0.20;0.18]
QIDS	T1–T2	82	6.23	0.51	212	12.21	<.001	1.15 [0.93;1.33]
QIDS	T2–T3	35	0.00	0.71	219	0.0	.99	0.01 [−0.23;0.26]
QIDS	T1–T3	29	6.23	0.72	221	8.61	<.001	1.16 [0.89;1.42]
QIDS	Treat vs wait	82	6.25	0.89	212	7.02	<.001	1.16 [0.83;1.48]

Degrees of freedom method: Kenward-Roger.

A similar pattern was observed for depressive symptoms. The linear mixed-effects model indicated a significant effect of time on QIDS scores (F = 82.01, p < .001). The higher intraclass correlation coefficient (ICC = 0.39) suggested a greater contribution of stable between-patient differences relative to PCL-5 scores, alongside meaningful within-person change over time. Depressive symptoms showed no significant change during the waiting period (d = 0.01, p = .96), whereas a large reduction was observed during the treatment phase (d = 1.15, p < .001). The contrast comparing symptom change during the waiting period and treatment phase was significant (d = 1.16, p < .001), indicating greater improvement during PE than during the waiting period. Treatment gains were maintained at follow-up relative to baseline (d = 1.16, p < .001) (see Tables 3 and 4).

Sensitivity analyses assessing the missing-at-random (MAR) assumption showed no significant differences between participants with and without follow-up data in PCL-5 scores at intake, baseline, post-treatment, or number of sessions (all p > .05).

Regarding clinically meaningful change, 79.79% of participants exhibited a reliable improvement, defined as a ≥ 10-points decrease on the PCL-5 from pre- to post-treatment. At post treatment, 75.53% of participants scored below the clinical cutoff of 33 on the PCL 5, consistent with probable remission of PTSD. compared to 12.63% at baseline (T1).

3.2. Exploratory analysis

An exploratory quadratic linear mixed-effects model (Figure 3) with random slopes for time was fitted to examine the trajectory of PTSD symptoms across sessions. The model intercept was 49.12 (SE = 1.48), indicating that the estimated mean PCL-5 score at session 1 was 49.12. The model revealed significant linear (β = −2.53, SE = 0.40, t = −6.27, p < .001) and quadratic effects of time (β = −0.18, SE = 0.02, t = −7.39, p < .001), indicating a non-linear decrease in PCL-5 scores over the course of treatment. The negative quadratic term suggests an accelerating reduction in symptoms across sessions. Random effects indicated substantial inter-individual variability in both baseline symptom severity (SD = 13.22) and rate of change over time (SD = 2.98), with a small correlation between intercept and slope (r = −0.09). Estimated marginal means showed a marked decrease in PTSD symptoms from session 1 to later sessions. Planned contrasts indicated large and statistically significant reductions between session 1 and session 8 (Δ = −26.66, 95% CI [−31.55, −21.76], p < .001, d = −1.94, 95% CI [−2.30, −1.58]) and between session 1 and session 12 (Δ = −49.89, 95% CI [−57.71, −42.08], p < .001, d = −3.63, 95% CI [−4.20, −3.06]). These results indicate substantial symptom reductions over time.

Figure 3.

Observed and model-estimated trajectories of PTSD symptoms (PCL-5 scores) across PE sessions.

3.3. Feasibility and safety

Only one participant (1%) showed a clinical exacerbation of PTSD symptoms (≥10-points increase on the PCL-5). Ten participants (9.47%) met criteria for dropout, defined as completing fewer than eight PE sessions while remaining symptomatic (PCL-5 ≥ 33).

4. Discussion

The present study evaluated the feasibility and preliminary effectiveness of PE in a French outpatient trauma clinic with implementation barrier. Consistent with existing literature, our results suggest a large and sustained reduction in PTSD symptom severity following PE, which exceeded the symptom changes observed during the waiting period. Specifically, during the pre-treatment interval (approximately four months), a small but statistically significant decline in PTSD symptoms was observed. Nevertheless, during an average of 8 PE sessions, patients exhibited a large reduction in PTSD symptoms significantly greater than the reduction from intake to baseline. Consistent with prior meta-analytic findings, PTSD symptoms change remained large at follow up (Kline et al., 2018). Although a small decrease in PTSD symptom scores was observed between endpoint and follow-up, this change did not reach statistical significance A similar pattern was observed for depressive symptoms (van Minnen et al., 2015), although the magnitude of improvement was larger than expected. Additionally, our results also showed that a large majority (80.85%) of the sample experienced clinically significant symptom reduction, and 75.52% scored below the clinical cutoff consistent with probable remission which is aligned with our hypothesis based on results reported by Milligan et al. (2025).

Further, only 9.47% of participants met our definition of dropout which was clinically grounded by considering both treatment dose (not receiving 8 sessions) and symptomatology level (still above PCL-5 cutoff). Despite concerns from some researchers and clinicians (Najavits, 2015), these findings are reassuring and below those reported in prior controlled studies like RCTs on trauma focus psychotherapy in routine care that observed generally around 20% of dropout (e.g. Goetter et al., 2015; Varker et al., 2021). This result confirms our hypothesis. Only a small minority of patients who initiated PE discontinued treatment while remaining above the clinical cutoff for PTSD symptoms. Finally, only one participant (1%) experienced a clinical exacerbation of symptoms. Previous studies suggest that patients receiving trauma-focused therapies are not at greater risk of symptom worsening than patients on a waiting list (Ehlers et al., 2013; Jayawickreme et al., 2014; Murray et al., 2022). Our findings do not support concerns raised by some professionals that exposure therapy might be poorly tolerated, unethical, or potentially harmful to patients (Deacon et al., 2013).

Beyond the primary pre-post analyses, the exploratory trajectory models provided convergent evidence supporting substantial reductions in PTSD symptoms over the course of PE treatment. Whereas the main analyses estimated average symptom differences between fixed assessment points, the continuous-time mixed-effects models allowed the examination of symptom trajectories across treatment sessions while accounting for inter-individual variability in rates of change. The results suggested that PTSD symptom reduction was not strictly linear over time and that patients differed in their rates of improvement throughout treatment, highlighting the potential heterogeneity of treatment response in routine clinical care. Clinically, substantial reductions in PTSD symptoms were already observed around session 8, which corresponds closely to the average treatment duration in the present study and to the lower range of standard PE protocols typically delivered in routine care. Overall, these exploratory findings are consistent with the main analyses indicating large symptom reductions following treatment. However, the trajectory analyses should be interpreted cautiously. The quadratic model was exploratory, and the number of observations decreased at later sessions, particularly beyond session 8. Consequently, model-estimated reductions at later sessions, including the very large effect size estimated at session 12, may partly reflect extrapolation from the fitted model rather than precise empirical estimates.

Despite interesting results, several limitations should be acknowledged. This study was not preregistered. Given its retrospective and naturalistic design, results should be considered exploratory and interpreted with caution. The primary limitation of this study is its single-arm pre-post observational design. Without randomisation or a concurrent comparison condition, it is not possible to determine the extent to which symptom reductions were specifically attributable to PE rather than to spontaneous improvement, nonspecific therapeutic effects, concurrent interventions, regression to the mean, or repeated-assessment effects. Although the within-subject waiting-list phase provided a useful reference period, it did not constitute a strictly passive control condition, as participants could receive treatment as usual, including pharmacological management or supportive care, which were not systematically assessed in the present study. Consequently, some symptom improvement during the waiting period may reflect nonspecific therapeutic effects associated with clinical engagement (Cuijpers et al., 2019), as well as placebo and repeated-assessment effects commonly reported in PTSD trials (Motta et al., 2023). Findings should therefore be interpreted as preliminary evidence of feasibility and effectiveness in routine care rather than definitive evidence of treatment efficacy.

Moreover, the retrospective convenience sampling strategy represents an additional limitation. Because the study relied on accessible routine-care clinical records rather than a prospective recruitment procedure or systematic screening log, we could not determine the total number of eligible patients during the study period or compare included and non-included patients. Consequently, selection bias cannot be ruled out, and the sample may not be fully representative of all patients treated at the clinic. Furthermore, we lacked detailed demographic and clinical sample information, socio-economic status and standardised comorbidity measures, which limits our capacity to control for potential confounding variables. In particular, data on medication was not systematically collected, and co-occurring medication treatment may have contributed to symptom reduction. However, first line medications including antidepressants usually yield effect size (ISTSS guidelines, 2018) that are much smaller than the effect sizes we observed over the treatment phase in our sample. Another limitation is the absence of formal assessment of treatment fidelity (e.g. session recordings, adherence ratings, or standardised fidelity scales). Consequently, variability in the delivery of Prolonged Exposure across therapists cannot be ruled out, which may affect the interpretation of treatment effectiveness. In addition, therapist-level effects were not modelled (e.g. via random intercepts for therapists). As a result, variability attributable to therapist differences may have influenced outcomes. However, these limitations are consistent with the naturalistic conditions of routine clinical practice. Another limitation is follow-up assessments were missing for 61 of the 95 participants, and missingness may not have been completely at random and should be interpreted with considerable caution. However, comparisons of post-treatment PCL-5 scores between participants with missing follow-up data and completers suggested no evidence of systematic differences associated with missingness. Our sample may also not be representative of the broader PTSD population, as it only includes individuals seeking treatment in an outpatient setting, potentially excluding individuals with more severe avoidance or motivational barriers.

This study contributes to the growing practice-based evidence on PE by examining its implementation in a naturalistic French clinical context, where it had not previously been systematically evaluated. The findings may also support the recent national development of regional psychotrauma centres in France (El-Hage et al., 2019), which were intended to improve access to specialised trauma-focused care within the public mental health system. Because PE was delivered in a public hospital-based outpatient trauma clinic with relatively few exclusion criteria, the findings may be particularly relevant to other publicly funded healthcare systems seeking to implement trauma-focused evidence-based treatments under routine care conditions. Importantly, PE was successfully delivered by clinicians with heterogeneous therapeutic orientations and limited prior clinical experience who received relatively low-intensity implementation support consisting primarily of brief formal training and ongoing peer case consultation. Together with the relatively low dropout rate and substantial symptom reductions observed after an average of eight sessions, these findings suggest that PE may be feasibly implemented in routine outpatient settings even when specialised trauma-focused expertise is initially limited. These results are also consistent with international implementation research highlighting the adaptability of PE across diverse clinical systems and cultural contexts (Becker et al., 2004; McLean & Foa, 2013).

5. Conclusion

Our results suggest that PE may be feasibly implemented in routine outpatient care and may be associated with clinically meaningful symptom reductions within routine outpatient care in France. The substantial symptom reductions, together with low dropout and clinical exacerbation rates, support the safety and feasibility of PE in real-world clinical settings. Beyond its immediate context, these findings underscore the potential for successful implementation of PE within healthcare systems where evidence-based psychotherapies remain underutilised. Strengthening these practice-oriented efforts will be critical to expand the reach and equity of evidence-based trauma care in underserved populations and across diverse healthcare environments.

Disclosure statement

Dr. Bui reports speaking or consulting fees from Johnson & Johnson, licenses or royalties from Springer and Wolters Kluwyer. Mr. Rascol, Mr. Gosselin, Mrs. Avrillon, Mr. Brodin, Mrs Gomes Lopes report royalties from De Boeck Superieur.

Generative AI statement

ChatGPT (GPT 5.1; OpenAI) was used during the writing process to improve language and coding assistance. After using the tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

Data availability statement

The data supporting the findings of this study are not publicly available due to ethical and legal restrictions related to participant confidentiality.