Loss of PTSD Diagnosis in Response to Evidence-Based Treatments

Key Points

Question

What proportion of patients lose their posttraumatic stress disorder (PTSD) diagnosis after receipt of evidence-based psychotherapies?

Findings

This systematic review and meta-analysis included 34 randomized clinical trials and found that military and veteran (milvet) samples had lower proportions of diagnosis loss than nonmilvet samples for cognitive processing therapy and prolonged exposure. Exploratory analyses showed some evidence that eye movement desensitization and reprocessing had the highest proportion of diagnosis loss.

Meaning

These findings supplement existing data on these treatments and may be useful to clinicians when discussing treatment selection and progress with patients.

Abstract

Importance

In recent decades, evidence-based psychotherapies to treat posttraumatic stress disorder (PTSD) have been developed with robust evidence bases. However, efficacy observed in clinical trials is not always directly applicable to clinical practice.

Objective

To estimate the percentage of patients in both military and veteran (hereafter milvet) and nonmilvet populations that lose their PTSD diagnosis after treatment.

Data Sources

We used the PTSD Repository to identify studies with adults with a DSM-IV/DSM-5 PTSD diagnosis based on a validated assessment. The repository, maintained by the US National Center for PTSD, is continually updated with randomized clinical trials and includes studies published from January 1988 on.

Study Selection

For eligibility, PTSD had to be the primary treatment target, with psychotherapy applied as monotreatment. Eligible studies reported the number of participants who did not meet diagnostic criteria for PTSD posttreatment. When this review was initiated (October 2023), the repository contained 496 unique studies. Data analysis was completed from October 2023 to June 2024.

Data Extraction and Synthesis

The repository follows Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guidelines and uses Cochrane Risk of Bias 2.0. We used mixed-effects logistic regression models to estimate diagnosis loss and incorporated milvet status and sex as covariates.

Main Outcomes and Measures

The primary outcome was the proportion of participants who no longer met criteria for a diagnosis of PTSD posttreatment as assessed by a validated instrument.

Results

We included 34 randomized clinical trials (N = 3208 participants). Point estimates of diagnosis loss across trauma-focused treatments for nonmilvet samples ranged from 65% to 86%. Milvet samples had lower proportions of diagnosis loss in studies of cognitive processing therapy and prolonged exposure compared to nonmilvet samples, ranging from 44% to 50%. There was substantial overlap between the covariates of milvet status and sex. An exploratory analysis identified eye movement desensitization and reprocessing as having the highest proportion of diagnosis loss, but there was substantial heterogeneity, and none of the studies were milvet-focused or conducted in the US. Also, 95% confidence intervals partially overlapped for all trauma-focused treatment estimates.

Conclusions and Relevance

This systematic review and meta-analysis contributes to the substantial literature on psychotherapeutic treatments for PTSD by meta-analyzing the probabilities of diagnosis loss for each psychotherapy. Diagnosis loss data are a relatively straightforward way to discuss potential benefits when initiating a therapy or when discussing potential barriers to progress in treatment.

This systematic review and meta-analysis estimates the percentage of patients in both military and veteran and non–military and veteran populations that lose their posttraumatic stress disorder (PTSD) diagnosis after psychotherapeutic treatment.

Introduction

Posttraumatic stress disorder (PTSD) is a clinically significant condition characterized by emotional, physical, and behavioral symptoms that persist for more than 1 month after exposure to a traumatic event (ie, threatened or actual death, serious injury, or sexual violence).¹ Traumatic events take many forms, including combat exposure, interpersonal violence, and natural disasters. Prevalence rates in the general population vary. One survey estimated a lifetime prevalence of 6% and a 1-year prevalence of 5% in a nationally representative sample of the general population of US adults.² Prevalence rates in military and veteran samples tend to be higher. One study based on the 2019-2020 National Health and Resilience in Veterans Study found a weighted lifetime prevalence of 9.4% in US veterans.³ Judkins and colleagues⁴ examined annual incidence rates from 2001 to 2017 and found that PTSD diagnoses steadily climbed among active duty service members (ADSMs) from a low of 0.01% in 2002 to a high of 13% in 2016. Data collected between 2016 and 2020 found an average annual PTSD incidence rate of 9.6% among ADSMs.⁵

Over the last 4 decades, psychotherapies have been developed and adapted to address PTSD in both military and veteran (hereafter, milvet) and nonmilvet (individuals who have never served in the military) populations. Cognitive behavioral therapy (CBT), an evidence-based approach used for many types of disorders, has been adapted to treat PTSD by focusing on challenging thoughts, modifying behaviors, and using exposure techniques pertaining to traumatic experiences.⁶ Cognitive processing therapy (CPT), a manualized cognitive-behavioral treatment, focuses on challenging and restructuring particular types of unhelpful beliefs related to a specific traumatic event.^7,8 Prolonged exposure (PE) is a systematic approach to recalling trauma-related memories, feelings, and situations.^9,10 Eye movement desensitization and reprocessing (EMDR) involves processing trauma-related memories, thoughts, and feelings through a combination of exposure techniques and forms of sensory stimulation (eg, bilateral eye movements).^11,12 Cognitive therapy for PTSD (CT-PTSD) focuses on modifying negative appraisals of the trauma or its sequelae and restructuring unhelpful beliefs and behaviors that maintain PTSD.¹³ Among non–trauma-focused treatments, present-centered therapy (PCT) has frequently been used as an active control comparator in trials evaluating trauma-focused treatments.¹⁴

Although the literature base supporting treatments for PTSD is substantial,^{15,16,17,18,19,20} results are often reported in terms of differences between treatment and control groups. This does not readily translate into clinical practice, which is concerned with alleviating symptoms and, ultimately, diagnoses. In assessing treatment response, it is helpful for clinicians to know what proportion of patients typically improve after receiving a given therapy. Studies vary in how response to PTSD treatment is measured, with some linking it with loss of PTSD diagnosis,^21,22 others defining it in terms of “reliable change” or clinically significant improvement on an instrument,^23,24 and still others associating it with different definitions of remission. Remission has been considered to be the point at which diagnostic criteria were last met,^25,26 the point at which symptoms ended,²⁷ or in terms of predefined cutoffs on instruments like the Clinician-Administered PTSD Scale (CAPS)²⁶ or the PTSD Checklist for DSM-5.²⁸ The variety of methods for defining treatment response underscores the difficulty of identifying when someone has made meaningful clinical gains.

Knowing which evidence-based treatments are associated with the highest proportions of diagnosis loss could guide treatment selection (eg, which treatment a population typically responds to) and help interpret patient progress during treatment (eg, if the patient is not improving despite commitment to the tasks of therapy and clinician fidelity to the therapy, mismatch between patient and treatment is a possibility). Prior research has established estimates of diagnosis loss for trauma-focused treatments as a class (not for each individual treatment) and for a smaller range of protocols.²⁹ Therefore, the aim of this systematic review and meta-analysis was to estimate the percentage of patients in both milvet and nonmilvet populations that lose their PTSD diagnosis (per DSM-IV²⁷ or DSM-5¹ criteria) after assignment to or receipt of evidence-based treatment for PTSD.

Methods

Protocol and Search Strategy

We conducted a systematic review of randomized clinical trials (RCTs) that evaluated efficacy of evidence-based psychotherapies for PTSD and reported the percentage of patients who lost their PTSD diagnosis at posttreatment. A protocol was not prepared and the study was not registered, as our data source was the PTSD Repository, a publicly available database created and maintained by the US National Center for PTSD. This repository is continually updated and contains RCTs of PTSD treatments published beginning January 1, 1988. It has been used in other systematic reviews.^29,30,31 To identify studies for inclusion in the repository, 8 databases are searched (Ovid, MEDLINE, PsycINFO, Cochrane CENTRAL, Embase, CINAHL, Scopus, and PTSDpubs) and then articles are dually reviewed. Studies are eligible if they were conducted among adults diagnosed with PTSD with a validated instrument, any kind of PTSD-specific treatment was applied, and any PTSD outcome was reported.³² At the start of this systematic review (October 2023), 496 RCTs were included in the repository.

Eligibility Criteria, Screening, and Data Extraction

When studies are added to the repository, data are extracted and made available to researchers. The research team reviewed all available RCTs and applied the following additional inclusion criteria: study reported the number of individuals who met or did not meet PTSD diagnostic criteria pretreatment and posttreatment; authors used an intention-to-treat analysis; treatment was psychotherapy alone; and the sample was recruited to treat PTSD as the primary diagnosis. A full accounting of exclusions can be found in Figure 1. The first 2 authors (T.M. and D.S.) verified diagnosis loss data used in the analyses by reviewing the original studies. The team used descriptions of treatments and sample-level characteristics (eg, sex) from study-level details in the repository.

Figure 1. Accounting of Exclusions From the Posttraumatic Stress Disorder (PTSD) Repository Database.

Study Quality Assessment

The PTSD Repository includes risk of bias assessments using Cochrane’s Risk of Bias 2.0 system.³³ Because this review focused on within-group changes rather than between-group differences, not all domains contained in Cochrane’s Risk of Bias assessment were relevant. Studies were judged based upon masking of clinicians, masking of participants, masking of assessors, and outcome attrition. We did not assess risk of bias arising from reporting bias due to an insufficient number of studies in each treatment type. We did not complete a certainty assessment, as the body of evidence has been assessed in prior meta-analyses.^15,17,19,34

Statistical Analysis

Primary data analysis was conducted between October 2023 and June 2024. We used mixed-effects logistic regression models to estimate diagnosis loss. The number of treatment group participants who lost the diagnosis was specified as the outcome variable in the models. Our primary analysis assumed data missing at random and used the total number who provided data at posttreatment in the denominator. A secondary analysis assumed data missing not at random and used the total number assigned to treatment in the denominator. We estimated the models separately by therapy protocol, combining treatment arms within a study that used the same protocol. We also evaluated 2 demographic factors as covariates: (1) sample was at least 80% milvet and (2) sample was at least 80% female (predominantly female). We report the summary proportions with diagnosis loss according to both missing data assumptions, 95% confidence intervals, and τ² estimates of between-study heterogeneity. For protocols that allowed for the inclusion of covariates, we also report the odds ratio and 95% confidence intervals for comparisons between the categories. We used a leave-one-out analysis across the treatment arms within each therapy protocol to assess the stability of the summary estimates. To account for other sources of heterogeneity (ie, risk of bias in studies and different definitions of diagnosis loss), we performed additional analyses that confirmed our primary analyses (see the eAppendix in Supplement 1). We estimated all models in Stata version 16 (StataCorp).³⁵

Results

A total of 34 RCTs were included in the analyses (N = 3208 participants). PE and CBT were the most frequently reported active therapy protocols and were used in 9 trial arms each. CPT was used in 6 trial arms, while EMDR and PCT were present in 5 each. Three trial arms used CT-PTSD, and all 3 were from the same lead investigator.^36,37,38 Finally, there were 16 waitlist or minimal attention (WL/MA) trial arms with diagnosis loss estimates. CAPS was the most common outcome assessment. Table 1 provides the descriptive information from the included studies, stratified by therapy protocol.^{20,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64} Risk of bias ratings indicated some or high concerns for most of the RCTs (Table 2).^{20,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68} There was no masking of treatment allocation to study participants or of treatment protocol to clinicians in any of the trial arms. Some studies used a masked assessor, but a substantial risk of bias remains, as the measures used to ascertain PTSD symptoms relied on self-disclosure from unmasked participants.

Table 1. Descriptive Information for the Included Study Treatment Groups by Treatment Protocol.

Source	Measure	PTSD diagnosis loss definition	Milvet	≥80% Female	No. assigned/No. posttreatment	% Loss (95% CI)
						All assigned participants	Participants available at post
Cognitive behavioral therapy
Bisson et al,39 2022	NS	NMC	No	No	99/83	72 (62-80)	86 (76-92)
Bryant et al,40 2003	CAPS	NMC	No	No	20/15	65 (41-85)	87 (60-98)
Castillo et al,41 2016	CAPS	Total score <45	Yes	Yes	44/34	50 (35-65)	65 (46-80)
Dunne et al,42 2012	SCID	NMC	No	No	13/12	62 (32-86)	67 (35-90)
Efendi et al,43 2020	CAPS	NMC	No	No	45/45	82 (68-92)	82 (68-92)
McDonagh et al,44 2005	CAPS	NMC	No	Yes	29/17	28 (13-47)	47 (23-72)
Mueser et al,45 2008	CAPS	NMC	No	No	54/32	20 (11-34)	34 (19-53)
Mueser et al,46 2015	CAPS	NMC	No	No	104/86	30 (21-40)	36 (26-47)
Sloan et al,20 2018	NS	NMC	Yes	No	98/74	22 (15-32)	30 (20-41)
Cognitive therapy for PTSD
Ehlers et al,36 2003	PDS	NMC; score <14	No	Unk	28/28	79 (59-92)	79 (59-92)
Ehlers et al,37 2005	CAPS	NMC and PDS score <14	No	No	14/14	71 (42-92)	71 (42-92)
Ehlers et al,38 2014	NS	NMC	No	No	61/61	75 (63-86)	75 (63-86)
Cognitive processing therapy
Angelakis et al,47 2020	CAPS	NMC	No	Yes	18/18	67 (41-87)	67 (41-87)
Monson et al,48 2006	CAPS	NMC	Yes	No	30/24	40 (23-59)	50 (29-71)
Pigeon et al,49 2022	CAPS	NMC	No	Yes	110/61	35 (26-44)	62 (49-74)
Resick et al,50 2002	CAPS	NMC	No	Yes	62/41	53 (40-66)	80 (65-91)
Resick et al,51 2017	PSS-I	NMC	Yes	No	268/165	27 (22-33)	44 (36-52)
Schnurr et al,52 2022	CAPS	NMC; ≥10-point improvement; severity <25	Yes	No	461/298	28 (24-33)	44 (38-49)
Eye movement desensitization and reprocessing therapy
Acarturk et al,53 2016	MINI	NMC	No	No	49/37	61 (46-75)	81 (65-92)
Nijdam et al,54 2012	SI-PTSD	NMC	No	No	70/48	64 (52-75)	94 (83-99)
Sack et al,55 2016	SCID	NMC	No	No	139/116	80 (72-86)	96 (90-99)
van den Berg et al,56 2015	CAPS	NMC	No	No	55/44	60 (46-73)	75 (60-87)
van Vliet et al,57 2021	CAPS	NMC	No	No	121/98	55 (45-64)	67 (57-76)
Present-centered therapy
Foa et al,58 2018	PSS-I	NMC	Yes	No	110/88	32 (23-41)	40 (29-51)
Lely et al,59 2019	CAPS	Total score <40 and ≥10-point decrease	No	No	15/12	13 (2-40)	17 (2-48)
McDonagh et al,44 2005	CAPS	NMC	No	Yes	22/20	32 (14-55)	35 (15-59)
Schnurr et al,60 2007	CAPS	NMC; total score <45	Yes	Yes	143/126	20 (14-28)	23 (16-31)
Sloan et al,20 2018	NS	NMC	Yes	No	100/88	20 (13-29)	23 (14-33)
Prolonged exposure
Foa et al,61 1999	PSS-I	NMC	No	Yes	25/23	60 (39-79)	65 (43-84)
Foa et al,58 2018	PSS-I	NMC	Yes	No	220/172	32 (26-38)	41 (33-48)
Morland et al,62 2019	CAPS	NMC	Yes	No	175/125	41 (33-48)	57 (48-66)
Peterson et al,63 2023	CAPS	NMC	No	No	234/138	32 (27-39)	55 (46-64)
Resick et al,50 2002	CAPS	NMC	No	Yes	62/40	53 (40-66)	82 (67-93)
Schnurr et al,60 2007	CAPS	NMC; score <45	Yes	Yes	141/120	39 (31-48)	46 (37-55)
Schnurr et al,52 2022	CAPS	NMC; ≥10-point improvement; severity <25	Yes	No	455/300	40 (36-45)	61 (56-67)
van den Berg et al,56 2015	CAPS	NMC	No	No	53/47	57 (42-70)	64 (49-77)
Yuen et al,64 2015	CAPS	NMC	Yes	No	74/52	31 (21-43)	44 (30-59)
Waitlist/minimal attention
Acarturk et al,53 2016	MINI	NMC	No	No	49/33	6 (1-17)	9 (2-24)
Dunne et al,42 2012	SCID	NMC	No	No	13/11	8 (0-36)	9 (0-41)
Efendi et al,43 2020	CAPS	NMC	No	No	45/45	11 (4-24)	11 (4-24)
Ehlers et al,36 2003	PDS	NMC and PDS score <14	No	Unk	29/27	28 (13-47)	30 (14-50)
Ehlers et al,37 2005	CAPS	NMC and PDS score <14	No	No	14/14	0 (0-23)	0 (0-23)
Ehlers et al,38 2014	NS	NMC	No	No	30/30	7 (1-22)	7 (1-22)
Falsetti et al,65 2008	CAPS	NMC	No	Yes	31/23	26 (12-45)	35 (16-57)
Foa et al,61 1999	PSS-I	NMC	No	Yes	15/15	0 (0-22)	0 (0-22)
Foa et al,58 2018	PSS-I	NMC	Yes	No	40/39	20 (9-36)	21 (9-36)
Krupnick et al,66 2008	CAPS	NMC	No	Yes	16/7	13 (2-38)	29 (4-71)
Lindauer et al,67 2005	SI-PTSD	NMC	No	No	12/12	25 (5-57)	25 (5-57)
McDonagh et al,44 2005	CAPS	NMC	No	Yes	23/20	17 (5-39)	20 (6-44)
Monson et al,48 2006	CAPS	NMC	Yes	No	30/27	3 (0-17)	4 (0-19)
Resick et al,50 2002	CAPS	NMC	No	Yes	47/40	2 (0-11)	3 (0-13)
Sloan et al,68 2012	CAPS	NMC	No	No	24/24	13 (3-32)	13 (3-32)
van den Berg et al,56 2015	CAPS	NMC	No	No	47/39	28 (16-43)	33 (19-50)

Abbreviations: CAPS, Clinician-Administered PTSD Scale; milvet, 80% or more military/veteran sample; MINI, Mini-International Neuropsychiatric Interview; NMC, not meeting criteria (ie, no longer met PTSD diagnostic criteria following treatment); NS, not specified; PDS, Posttraumatic Diagnostic Scale; PSS-I, PTSD Symptom Scale – Interview; PTSD, posttraumatic stress disorder; SCID, Structured Clinical Interview for DSM Disorders; SI-PTSD, Structured Interview for PTSD; Unk, unknown.

Table 2. Selected Domains From the Study-Level Risk of Bias Assessment.

Source	Participant unmasked	Clinician unmasked	Assessor unmasked	Low outcome attrition	Summary risk of bias
Acarturk et al,53 2016	Yes	Yes	No	No	Some concerns
Angelakis et al,47 2020	Yes	No	No	Yes	Low
Bisson et al,39 2022	Yes	Yes	No	Yes	Low
Bryant et al,40 2003	Yes	Yes	No	No	Some concerns
Castillo et al,41 2016	Yes	Yes	No	No	Some concerns
Dunne et al,42 2012	Yes	Yes	Probably yes	Yes	High
Efendi et al,43 2020	Yes	Yes	No information	Probably yes	High
Ehlers et al,36 2003	Yes	Yes	No	Yes	Some concerns
Ehlers et al,37 2005	Yes	Probably yes	Yes	Yes	High
Ehlers et al,38 2014	Yes	Yes	No	Yes	High
Falsetti et al,65 2008	Yes	Yes	No	No	High
Foa et al,61 1999	Yes	Yes	No	Yes	Some concerns
Foa et al,58 2018	Yes	Yes	No	Yes	Some concerns
Krupnick et al,66 2008	Yes	Yes	No information	No	High
Lely et al,59 2019	Yes	Yes	No	Yes	Some concerns
Lindauer et al,67 2005	Yes	Yes	No	No	High
McDonagh et al,44 2005	Yes	Yes	No	No	High
Monson et al,48 2006	Yes	Yes	No	Yes	Low
Morland et al,62 2019	Yes	Yes	No	No	Some concerns
Mueser et al,45 2008	Yes	Yes	No	No	Some concerns
Mueser et al,46 2015	Yes	Yes	No	Yes	Low
Nijdam et al,54 2012	Yes	Yes	Yes	No	High
Peterson et al,63 2023	Yes	Yes	Yes	No	High
Pigeon et al,49 2022	Yes	Yes	No	No	High
Resick et al,50 2002	Yes	Yes	No	No	High
Resick et al,51 2017	Yes	Yes	Yes	No	Some concerns
Sack et al,55 2016	Yes	Yes	No information	Yes	Some concerns
Schnurr et al,60 2007	Yes	Yes	No	Yes	Low
Schnurr et al,52 2022	Yes	Yes	No	No	Some concerns
Sloan et al,68 2012	Yes	Yes	No	Yes	Low
Sloan et al,20 2018	Yes	Yes	No	Yes	Low
van den Berg et al,56 2015	Yes	Yes	Yes	Yes	High
van Vliet et al,57 2021	Yes	Yes	No information	No	High
Yuen et al,64 2015	Yes	Yes	Yes	No	High

The summary estimates and subgroup comparisons, where available, are displayed in Table 3 and presented by missing data assumption. Here, we focus on the analyses for participants available at posttreatment vs all assigned, since those at posttreatment are more likely to have received a sufficient treatment dose. The pattern of results was similar for both model assumptions, with diagnosis loss estimates for all assigned participants being smaller in magnitude. The summary estimates for CBT, CPT, EMDR, and PE showed substantial statistical heterogeneity. Milvet status explained the heterogeneity for CPT and partially explained the heterogeneity for PE, with milvet samples having lower proportions of diagnosis loss. Neither covariate accounted for the heterogeneity for CBT, which may be due to insufficient data for the subgroups. There was no variation in milvet status or sex for EMDR to account for the observed heterogeneity. There was no meaningful difference between covariate subgroups for CT-PTSD, PCT, or WL/MA.

Table 3. Summary Estimates of Loss of Diagnosis by Therapy Protocol, Missing Data Assumption, and Level of Covariate.

Therapy	Studies, No.	All participants at posttreatment					All participants assigned
		No.	% Diagnosis loss (95% CI)	τ2	I2, %	OR (95% CI)	No.	% Diagnosis loss (95% CI)	τ2	I2, %	OR (95% CI)
CBT	9	398	61 (44-76)	0.94	88	NA	506	47 (31-63)	0.91	91	NA
Nonmilvet sample	7	290	65 (47-80)	NA		1 [Ref]	364	51 (33-68)	NA	NA	1 [Ref]
Milvet sample	2	108	46 (18-77)	0.84	89	0.46 (0.10-2.15)	142	35 (12-66)	0.83	88	0.51 (0.11-2.30)
<80% Female sample	7	347	62 (43-78)	NA	NA	1 [Ref]	433	49 (32-67)	NA	NA	1 [Ref]
≥80% Female sample	2	51	57 (23-85)	0.94	91	0.80 (0.15-4.26)	73	38 (13-71)	0.89	91	0.63 (0.13-3.09)
CPTa	6	607	57 (45-68)	0.27	83	NA	949	38 (29-49)	0.22	85	NA
Nonmilvet/≥80% female sample	3	120	69 (60-77)	NA	NA	1 [Ref]	190	45 (34-57)	NA	NA	1 [Ref]
Milvet/<80% female sample	3	487	44 (40-48)	0.00	0	0.35 (0.23-0.54)	759	29 (23-36)	0.03	25	0.49 (0.30-0.78)
CT-PTSDb	3	103	76 (67-83)	0.04	16	NA	103	76 (67-83)	0.04	16	NA
EMDR	5	343	86 (72-93)	0.75	85	NA	434	65 (56-73)	0.14	85	NA
PCT	5	334	28 (21-36)	0.08	48	NA	390	24 (19-30)	0.04	31	NA
Nonmilvet sample	2	32	27 (14-48)	NA	NA	1 [Ref]	37	24 (12-41)	NA	NA	1 [Ref]
Milvet sample	3	302	28 (20-37)	0.08	17	1.03 (0.39-2.71)	353	24 (18-30)	0.04	9	0.99 (0.42-2.35)
<80% Female sample	3	188	29 (21-40)	NA	NA	1 [Ref]	225	25 (19-33)	NA	NA	1 [Ref]
≥80% Female sample	2	146	26 (17-38)	0.06	40	0.83 (0.40-1.75)	165	22 (15-31)	0.02	19	0.86 (0.46-1.61)
PE	9	1017	56 (48-64)	0.18	81	NA	1439	41 (35-46)	0.08	74	NA
Nonmilvet sample	4	248	66 (55-75)	NA	NA	1 [Ref]	374	47 (38-56)	NA	NA	1 [Ref]
Milvet sample	5	769	50 (42-58)	0.10	63	0.53 (0.31-0.92)	1065	37 (31-43)	0.07	53	0.66 (0.42-1.06)
<80% Female sample	6	834	54 (45-62)	NA	NA	NA	1211	38 (32-43)	NA	NA	1 [Ref]
≥80% Female sample	3	183	63 (49-76)	0.16	87	1.48 (0.73-2.97)	228	48 (38-58)	0.05	69	1.50 (0.95-2.37)
WL/MA	16	406	13 (8-20)	0.60	63	NA	465	11 (7-17)	0.49	59	NA
Nonmilvet sample	14	340	13 (8-21)	NA	NA	1 [Ref]	395	11 (7-18)	NA	NA	1 [Ref]
Milvet sample	2	66	11 (3-32)	0.59	63	0.79 (0.19-3.32)	70	10 (3-29)	0.49	58	0.90 (0.23,3.47)
<80% Female sample	10	274	12 (6-20)	NA	NA	1 [Ref]	304	11 (6-18)	NA	NA	1 [Ref]
≥80% Female sample	5	105	13 (5-27)	0.59	63	1.14 (0.37-3.47)	132	10 (4-21)	0.47	58	0.91 (0.32-2.54)

Abbreviations: CBT, cognitive behavioral therapy; CPT, cognitive processing therapy; CT-PTSD, cognitive therapy for posttraumatic stress disorder; EMDR, eye movement desensitization and reprocessing; milvet, 80% or more participants were military/veteran; NA, not applicable; OR, odds ratio; PCT, present-centered therapy; PE, prolonged exposure; Ref, reference; WL/MA, waitlist/minimal attention.

^a The 6 trial arms fall into the same 2 groups of 3 based on sex composition, so these variables are completely confounded.

^b No dropout was reported for the included trial arms, so the analytic result is the same irrespective of missing data assumption.

Point estimates of diagnosis loss with nonmilvet samples ranged from 65% to 86%. EMDR had the highest point estimate of 86% (95% CI, 72%-93%) and our exploratory analysis (eAppendix in Supplement 1) showed some evidence that EMDR’s proportion of diagnosis loss was superior to other treatment arms. However, in that analysis, the EMDR studies differed from the others in important ways (no milvet samples, all conducted outside of the US), and none had a low risk of bias determination. In addition, all nonmilvet treatment arms (excluding PCT and WL/MA) had overlapping 95% confidence intervals, indicating that the true value for any given treatment could be equivalent to another. Removal of the study by Sack and colleagues⁵⁵ in the leave-one-out analysis lowered the overall estimate for EMDR (80%; 95% CI, 67%-89%; τ² = 0.33; I² = 72%) and reduced statistical heterogeneity.

For CPT and PE with milvet samples, diagnosis loss point estimates ranged from 44% to 50% (Figure 2). For the remainder of the treatment arms, overall estimates are the most appropriate to interpret. CT-PTSD was associated with 76% diagnosis loss (95% CI, 67%-83%); CBT with 61% (95% CI, 44%-76%); PCT with 28% (95% CI, 21%-36%); and WL/MA with 13% (95% CI, 8%-20%). The removal of the 2018 study by Foa and colleagues⁵⁸ from the PCT analysis resulted in a summary estimate of 24% (95% CI, 19%-29%; τ² = 0.00; I² = 0%).

Figure 2. Percentage of Posttraumatic Stress Disorder (PTSD) Diagnosis Loss by Psychotherapy Type and Data Assumption.

Figure represents overall estimates (A) and moderated estimates (B). Error bars represent 95% confidence intervals. CPT, cognitive processing therapy; CT-PTSD, cognitive therapy for PTSD; EMDR, eye movement desensitization and reprocessing; milvet, 80% or more military/veteran sample; PCT, present-centered therapy; PE, prolonged exposure; CBT, cognitive behavioral therapy; WL/MA, waitlist/minimal attention.

For PE and CBT, we were able to incorporate sex as a covariate, but the overlap between milvet status and sex limited our ability to determine which covariate explained the results. Results differed between the 2 treatments: for PE, samples that were predominantly female had a higher percentage of diagnosis loss, while conversely, for CBT, predominantly female samples had a lower percentage of diagnosis loss.

Discussion

In this systematic review and meta-analysis, we estimated proportion of diagnosis loss in patients receiving evidence-based psychotherapies. We found that all active treatments performed better than WL/MA conditions, although the point estimates of diagnosis loss significantly varied by patient population. For nonmilvet samples, EMDR and CT-PTSD showed the highest rates of diagnosis loss at 86% and 75%, respectively. Milvet samples showed lower proportions of diagnosis loss than nonmilvet samples for protocols in which this analysis was possible (CPT and PE). PCT, a nontrauma, manualized treatment, showed the lowest proportion of diagnosis loss (28%).

By evaluating diagnosis loss as an outcome, our findings supplement previous reviews evaluating the efficacy of PTSD psychotherapies.^{14,15,18,19,69} While our findings indicate possible superiority of EMDR with respect to diagnosis loss, this warrants further research. None of the included EMDR studies were conducted in milvet populations or in the US. These population differences could account for these findings. Prior meta-analytic findings about the magnitude of effects associated with EMDR compared to other PTSD treatments are mixed. Consistent with our findings, a meta-analysis of exposure psychotherapies found that EMDR produced large effects, while PE yielded medium effects.¹⁸ However, 1 network meta-analysis of PTSD treatments¹⁵ found that CPT, PE, and EMDR outperformed WL/MA control conditions but were not different from each other. More research is needed to examine how EMDR compares to other PTSD treatments. This may be particularly important in milvet populations, as most PTSD treatments demonstrate moderate efficacy in these populations.⁷⁰

Our finding of lower rates of diagnosis loss in milvet samples in PE and CPT is consistent with other research that found larger effect sizes for PE in nonmilvet vs military populations.¹⁸ This review suggests that there may be a similar pattern with CBT. There was substantial overlap between milvet and sex composition, with almost all milvet studies being predominantly male. As such, we cannot fully disentangle the 2 variables in explaining the statistical heterogeneity. A prior meta-analysis⁷¹ indicated that women respond better to PTSD treatment, so it is possible that both covariates explain the results.

When treatment is initiated, the principles of shared decision-making emphasize the importance of collaborative discussion between the patient and clinician to discuss treatment options.³⁴ Many factors influence which treatment is pursued, including previous patient experiences, clinician competence, evidence regarding the treatment’s effectiveness, and patient preference. Diagnosis loss data are a relatively straightforward way to communicate potential benefits of a given treatment to aid in this discussion.²⁹ Moreover, the findings of this study suggest that some treatments may be preferable over others given the patient population. In terms of treatment response, measuring progress throughout treatment is encouraged to ensure patients are benefitting.⁷² If it becomes apparent that patients are not improving or declining, any barriers to treatment should be explored and addressed. Should a component of the therapy itself be a barrier (eg, reticence to undergo exposure), an alternative therapy may be the solution. Again, diagnosis loss data can aid in the discussion of other therapies to pursue.

Limitations and Future Directions

We only included studies identified in the PTSD Repository and were limited by the methodological decisions used to generate the database. Studies of pharmacotherapies for PTSD and joint approaches were outside the scope of this review. Our included studies did not incorporate populations common in general clinical practice (eg, those with elevated suicidality and comorbid substance use issues), and nearly all offered treatment in an individual rather than group format, limiting clinical applicability. The therapies we included are subject to their own inherent limitations. CPT, PE, and EMDR can be resource-intensive to implement, and PE and CPT have low clinician adoption rates despite robust dissemination efforts.⁷³ Multiple reviews have found elevated dropout rates with trauma-focused treatment, and there is some evidence that this is due to patient reticence or inability to recall specifics of their traumas.^74,75,76,77

We also limited our dataset to studies that reported on diagnosis loss and did not use other types of data (eg, clinically significant improvement) in our estimates. Most studies relied on the CAPS as a diagnostic interview, but there was variability between studies about scoring methods used and what constituted diagnosis loss. Diagnosis loss itself is an imperfect measure of treatment response, being the most liberal indicator of progress.⁷⁸ It was used in this study to reduce heterogeneity in definitions of treatment response, not to indicate that a client no longer warrants clinical engagement for PTSD or for other important clinical concerns (eg, quality of life). Future studies could examine other types of treatment approaches to PTSD (eg, other psychotherapies, joint therapies, or pharmacotherapies), with different populations, and using other definitions of treatment response. In addition, it would be useful to examine whether and how symptom severity at baseline impacts results. Finally, a measure of treatment fidelity in future studies would enhance findings.

The results of this study illuminate additional lines of inquiry. First, it would be beneficial to better understand why studies with milvet populations produce smaller effects compared to studies with nonmilvet samples. Several reasons for this discrepancy have been posited, including different types of trauma exposure (eg, combat vs assault), sample characteristics (fewer women in military samples), and aspects of military culture (prolonged deployments, mental health stigma), but none have been thoroughly investigated and all warrant further study.^79,80 There are some indications that complex PTSD (cPTSD) may be more prevalent than PTSD in milvet populations.⁸¹ Treatment for cPTSD may necessitate psychotherapeutic strategies beyond those for PTSD.⁸² Additional research is needed to determine if cPTSD is more prevalent in milvet populations and how this compares to nonmilvet prevalence.

Methodological decisions may also partially explain the differences between milvet and nonmilvet samples, as they are typically examined in separate studies. Studies in which both populations are incorporated could help mitigate this issue and may be feasible in military clinics that treat both ADSMs and their family members. In addition, CT-PTSD would benefit from independent replication,⁸³ as all studies came from the same author. Finally, an examination of EMDR with milvet samples would provide additional evidence regarding population differences in diagnosis loss.

Conclusions

The results of this systematic review and meta-analysis supplement the current literature on PTSD treatments and indicate that each of the trauma-focused treatments resulted in higher proportions of diagnosis loss than PCT or WL/MA. However, there were meaningful differences between populations. EMDR and CT-PTSD showed the highest rates of diagnosis loss, although neither included studies with milvet samples. CPT and PE showed significantly lower rates of diagnosis loss in milvet samples compared to nonmilvet samples. CBT showed similar trends in terms of milvet vs nonmilvet populations but was unique in that predominantly female samples responded less well than samples that were less than 80% female. Diagnostic loss data are a relatively straightforward way to communicate potential treatment benefits with patients. Thus, these data may be useful to clinicians and patients when selecting treatments and assessing barriers to progress. The differences seen in terms of milvet status indicate opportunities to better understand the reasons underlying these differences and to examine how EMDR and CT-PTSD perform with milvet samples.

Supplement 1.

eAppendix. Omnibus Mixed-Effects Logistic Regression Models With Additional Covariates: Risk of Bias and Outcome Definition

eTable 1. Freely Estimated and Constrained Coefficients From Omnibus Mixed-Effects Logistic Regression Model: Only Participants Available at Posttreatment

eTable 2. Freely Estimated and Constrained Coefficients From Omnibus Mixed-Effects Logistic Regression Model: All Participants Assigned

Supplement 2.

Data Sharing Statement