Linguistic markers for identifying post-traumatic stress disorder and associated symptoms: a systematic literature review

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Since the late 1990s, there has been growing interest in the use of linguistic markers in psychiatry, with advancements in natural language processing contributing to this development.

“Since the late 1990s, there has been a growing interest in the use of linguistic markers in psychiatry, with advancements in natural language processing (NLP) contributing to this development.[8][9][10]”

• cites [8] Wu, 2023, Systematic review of machine learning in PTSD studies for automated diagnosis evaluation, NPJ Ment Health Res sourcePartly supported

  The Results (Structured clinical interviews and Self-report questionnaires and narratives; page numbers unavailable) describe the use of speech, lexical and semantic features, including NLP, to assess PTSD. The Discussion describes recent advances in NLP models. The paper does not establish that interest in linguistic markers across psychiatry has grown since the late 1990s or that NLP advances drove that historical trend. (LLM, conf=0.83)

  Experimental — accuracy not yet measured
• cites [9] Zhang, 2022Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [10] Abbe, 2016, Text mining applications in psychiatry: a systematic literature review, Int J Methods Psychiatr Res sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Data extraction was conducted independently by four reviewers following the CHARMS checklist.

“Data extraction was conducted independently by 4 reviewers (R.Q., F.G, Y.A., and S.M.), following the recommendations in the CHARMS checklist.23”

• cites [23] Moons, 2014, Critical appraisal and data extraction for systematic reviews of prediction modelling studies: the CHARMS checklist, PLoS Med sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

PTSD is a severe mental health condition triggered by a traumatic event and characterized by intrusive memories, avoidance behaviors, negative changes in thought and mood, and alterations in arousal and reactivity.

“Post-traumatic stress disorder (PTSD) is a severe mental health condition triggered by experiencing a traumatic event and characterized by symptoms like intrusive memories, avoidance behaviors, negative changes in thought and mood, and alterations in arousal and reactivity.”

• cites [1] American Psychiatric Association sourceCould not verify

  The supplied passages (11 September 2026, no page given) discuss depression, not PTSD. Searches for PTSD and its proposed symptom clusters returned no passage addressing the claim, so it cannot be verified from our copy. (LLM, conf=0.76)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

Over 80% of PTSD cases involve one or more comorbid disorders such as depression, anxiety, or mild traumatic brain injury.

“A review 5 of the literature revealed that over 80% of PTSD cases involved 1 or more comorbid disorders such as depression, anxiety, or mild traumatic brain injury (TBI).6”

• cites [5] Bryant, 2019, Post-traumatic stress disorder: a state-of-the-art review of evidence and challenges, World Psychiatry sourcePartly supported

  The PREVALENCE section says epidemiological studies suggest that most people with PTSD have comorbid disorders, particularly depression and anxiety disorders. It does not give an over-80% figure or identify mild traumatic brain injury as an example; it names substance use disorder instead. No page number is available in our copy. (LLM, conf=0.88)

  Experimental — accuracy not yet measured
• cites [6] Miao, 2018, Posttraumatic stress disorder: from diagnosis to prevention, Mil Med Res sourcePartly supported

  The “Prevalence and importance” section says that more than 80% of PTSD patients have one or more comorbidities and identifies depression and concurrent mild TBI. It does not identify anxiety as a comorbidity; the “Definition and differential diagnosis” section mentions anxiety disorder only as a condition with similar symptoms to distinguish from PTSD. No page numbers are available in our copy. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

May combine two claims.

PROBAST was augmented with criteria tailored to machine-learning challenges: interpretability, performance metrics, code availability, and overfitting mitigation.

“To align with best practices 27 for systematic reviews of ML studies, we augmented PROBAST with criteria tailored to these challenges”

• cites [27] Collins, 2024, TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods, BMJ sourcePartly supported

  The Development of TRIPOD+AI section (page unavailable) describes PROBAST+AI as a tool being developed to assess the quality and risk of bias of machine-learning prediction models. The TRIPOD+AI statement section calls it forthcoming; it does not say PROBAST had already been augmented or identify the four listed criteria as PROBAST+AI criteria. Code, performance, and overfitting appear in the paper’s TRIPOD+AI reporting checklist instead. (LLM, conf=0.86)

  Experimental — accuracy not yet measured
• cites [28] Fryer, 2021, Shapley values for feature selection: the good, the bad, and the axioms, IEEE Access sourceCould not verify

  The Introduction and ‘Feature Selection in General’ sections discuss Shapley-based interpretation and model performance, not systematic-review criteria. Searches for PROBAST and for code availability and overfitting mitigation found no passage about augmenting PROBAST; the cited copy does not address the attributed point. Page numbers were not available in our copy. (LLM, conf=0.76)

  Experimental — accuracy not yet measured

May combine two claims.

Present tense usage often suggests that the traumatic experience is being relived rather than merely remembered, indicating ongoing distress and unresolved trauma.

“Present tense usage often suggests that the traumatic experience is being relived rather than merely remembered (related to criterion B of DSM-5), which can indicate ongoing distress and unresolved trauma.40,77”

• cites [40] Quillivic, 2024, Interdisciplinary approach to identify language markers for post-traumatic stress disorder using machine learning and deep learning, Sci Rep sourcePartly supported

  The Methods (Language features; page unavailable) report that prior studies observed increased present-tense use in PTSD narratives, but caution that present tense has several meanings depending on context. The Discussion (page unavailable) links possible literal re-experiencing under Criterion B to lexical features, disfluencies, and perception-related verbs—not specifically to present tense. It does not establish that present-tense use indicates ongoing distress or unresolved trauma. (LLM, conf=0.85)

  Experimental — accuracy not yet measured
• cites [77] Hellawell, 2004, A comparison of flashbacks and ordinary autobiographical memories of trauma: content and language, Behav Res Ther sourceNot checked — no full text of the source

May combine two claims.

Future tense was less commonly analyzed but has been considered to reflect anticipatory anxiety or fear about potential future threats.

“Meanwhile, the future tense was less commonly analyzed, yet has been considered to reflect anticipatory anxiety or fear about potential future threats, which are significant elements of PTSD symptomatology.40,59”

• cites [40] Quillivic, 2024, Interdisciplinary approach to identify language markers for post-traumatic stress disorder using machine learning and deep learning, Sci Rep sourcePartly supported

  The Results associate greater use of future tense with PTSD hypervigilance symptoms, and the Discussion suggests it may reflect altered anticipation of forthcoming activities. The paper does not say that future tense was less commonly analyzed or specifically interpret it as fear or anxiety about future threats. The supplied passages have no page numbers. (LLM, conf=0.82)

  Experimental — accuracy not yet measured
• cites [59] Manne, 2002, Language use and post-traumatic stress symptomatology in parents of pediatric cancer survivors, J Appl Soc Pyschol sourceSupported

May combine two claims.

Trauma narratives with a high density of negative emotion words were associated with poorer psychological adjustment and higher PTSD symptoms.

“Many others [55][56][57][58] also found that trauma narratives with a high density of negative emotion words were associated with poorer psychological adjustment and higher PTSD symptoms.”

• cites [55] Beaudreau, 2007, Are trauma narratives unique and do they predict psychological adjustment?, J Trauma Stress sourceNot checked — no full text of the source
• cites [56] Bartal, 2022 sourcePartly supported

  The Results (Word category analysis; page unavailable) report that women in the provisional childbirth-related PTSD group used more negative-emotion words in their birth narratives than women without childbirth-related PTSD. The paper does not establish a separate association with poorer psychological adjustment, and its finding concerns childbirth narratives and PTSD group status rather than trauma narratives generally or a graded relationship with symptom severity. (LLM, conf=0.85)

  Experimental — accuracy not yet measured
• cites [57] Ul Alam, 2020, LAXARY: sourcePartly supported

  The Related Works section reports more frequent negative-emotion words among depressed Twitter users and notes language differences in PTSD self-narratives. It does not say that a high density of negative-emotion words in trauma narratives was associated with poorer psychological adjustment or higher PTSD symptom levels; the paper's PTSD tweet analysis instead reports greater use of anxiety words. (LLM, conf=0.79)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

An increase in cognitive processing words could reflect heightened engagement with trauma content, potentially leading to more symptoms at first.

“Conversely, other research 41,48,74 indicated that an increase in cognitive processing words could reflect heightened engagement with trauma content, potentially leading to more symptoms at first.”

• cites [41] Fernández-Lansac, 2017, Quality of memories in women abused by their intimate partner: analysis of traumatic and nontraumatic narratives, J. Trauma. Stress sourceNot checked — no full text of the source
• cites [48] D’Andrea, 2012, Linguistic predictors of post-traumatic stress disorder symptoms following 11 September 2001, Appl Cogn Psychol sourcePartly supported

  The Interpretation of data section says greater use of cognitive-mechanism words shortly after 9/11 predicted greater hyperarousal symptoms months later, and suggests those words may reflect attempts to process trauma-related information. It measured greater use, not an increase in word use, and did not find more symptoms initially; it also offers emotional distancing as an alternative interpretation. (LLM, conf=0.85)

  Experimental — accuracy not yet measured
• cites [74] Booker, 2018, Narratives in the immediate aftermath of traumatic injury: markers of ongoing depressive and posttraumatic stress disorder symptoms, J Trauma Stress sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Study selection was crosschecked with existing literature reviews, and no additional studies were added.

“In order to avoid missing studies, we crosschecked our study selection with existing literature reviews 8,10,13,17,19 and no additional studies were added.”

• cites [8] Wu, 2023, Systematic review of machine learning in PTSD studies for automated diagnosis evaluation, NPJ Ment Health Res sourceSupported
• cites [10] Abbe, 2016, Text mining applications in psychiatry: a systematic literature review, Int J Methods Psychiatr Res sourceNot checked — no full text of the source
• cites [13] Zhang, 2022, Natural language processing applied to mental illness detection: a narrative review, NPJ Digit Med sourceSupported

May combine two claims.

Machine-learning models often involve higher complexity, greater risks of overfitting, and distinct interpretability needs compared with traditional regression.

“Unlike traditional regression, ML models often involve higher complexity (eg, non-linear algorithms, hyperparameter tuning), greater risks of overfitting, and distinct interpretability needs.”

• cites [26] Moriarty, 2022, Predicting relapse or recurrence of depression: systematic review of prognostic models, Br J Psychiatry sourcePartly supported

  The ROB and applicability assessment (p. 7) says machine-learning models may require substantially larger samples than regression models to achieve low optimism, and the Implications section (p. 8) identifies lack of interpretability as a criticism. The paper does not establish the claimed higher complexity or give non-linear algorithms and hyperparameter tuning as examples; it also discusses overfitting as a risk for statistical models generally rather than clearly establishing a greater risk for machine learning. (LLM, conf=0.85)

  Experimental — accuracy not yet measured

May combine two claims. The quote only partly states this claim.

Studies with the lowest risk-of-bias (RoB) score fell into two categories: clinical studies focusing on very simple linguistic features (e.g., narrative length or reading scores) or studies using manual annotators that increased the events-per-variable (EPV) number while maintaining gold standards; and studies involving multiple researchers from different departments or computer science teams, and

“The studies with the lowest RoB score fell into 2 categories: clinical studies focusing on very simple linguistic features, such as narrative length 85 or reading scores, 38 or studies using manual annotators, 77 which increased the EPV number while maintaining gold standards for diagnostic tools, participant inclusion, and linguistic features.”

• Also stated in: § Risk of bias · plain

  “The other type of study with low RoB involved multiple researchers from different departments to incorporate best practices from all disciplines (using gold standards for diagnosis and study design), along with robust and reliable statistical frameworks and linguistic analysis tools 56,60,86,87 or computer science teams 88 using publicly available datasets.”

• Also stated in: § Summary of findings · hedged

  “Studies with the lowest RoB often focused on simple linguistic characteristics or employed multidisciplinary teams using robust statistical methods.”

• cites [85] Fernández-Lansac, 2015, Narrative length and speech rate in battered women, PloS One sourceSupported
• cites [38] Gray, 2001, Complexity of trauma narratives as an index of fragmented memory in PTSD: a critical analysis, Appl Cogn Psychol sourceNot checked — no full text of the source
• cites [77] Hellawell, 2004, A comparison of flashbacks and ordinary autobiographical memories of trauma: content and language, Behav Res Ther sourceNot checked — no full text of the source

May combine two claims.

One study applied LIWC to translated data, introducing high risk of bias, while another applied it to a very small text of 50 words.

“One study 67 even applied LIWC to translated data, which introduced a high risk of bias (RoB), while another study 73 applied it to a very small text (50 words).”

• cites [67] Ng, 2015, Narrative characteristics of genocide testimonies predict posttraumatic stress disorder symptoms years later, Psychol Trauma sourcePartly supported

  The Measures section (p. 5) says this study used LIWC, and Strengths and Limitations (p. 8) says its LIWC measurements depended on translations from Kinyarwanda to English. The authors acknowledge possible measurement error but do not characterize it as a high risk of bias. The 50-word claim is attached to a different citation and is not assessed here. (LLM, conf=0.86)

  Experimental — accuracy not yet measured
• cites [73] Marshall, 2020, Prospective analysis of linguistic analysis as a method for assessing trauma symptoms after Hurricane Harvey among Houstonian adults, Traumatology sourceNot checked — no full text of the source

May combine two claims.

Studies with notably higher AUC scores include Reece et al, Kirinde Gamaarachchige and Inkpen, and Schultebraucks et al.

“From the forest plot highlights studies with notably higher AUC scores, particularly Reece et al, 52 Kirinde Gamaarachchige and Inkpen, 84 and Schultebraucks et al. 43”

• cites [52] Reece, 2017, Forecasting the onset and course of mental illness with Twitter data, Sci Rep sourceCould not verify

  The Results (Machine learning classifier; page unavailable) show an ROC curve and report precision, recall, specificity, and F1 scores, but our copy gives no AUC score for the Reece study. It therefore cannot verify the citing document’s characterization of that study as having a notably higher AUC in its forest plot. (LLM, conf=0.76)

  Experimental — accuracy not yet measured
• cites [84] Kirinde Gamaarachchige, 2019 sourceSupported
• cites [43] Schultebraucks, 2020, Deep learning-based classification of posttraumatic stress disorder and depression following trauma utilizing visual and auditory markers of arousal and mood, Psychol Med sourceNot checked — no full text of the source

May combine two claims.

The first two studies with higher AUC scores leverage social media data and rely on self-reported PTSD diagnoses, which may contribute to their higher AUC values.

“The first 2 studies leverage social media data and rely on selfreported PTSD diagnoses, which may contribute to their higher AUC values.”

• cites [52] Reece, 2017, Forecasting the onset and course of mental illness with Twitter data, Sci Rep sourceSupported
• cites [84] Kirinde Gamaarachchige, 2019 sourcePartly supported

  The paper’s Comparison to related work section reports an AUC above 0.90 for its PTSD classification, and the Conclusion identifies social media text as its data source (page numbers unavailable in our copy). The copy does not establish that its PTSD labels came from self-reported diagnoses or suggest that self-reporting contributed to the higher AUC. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

Interpretability tools ensure transparency, a priority emphasized in recent machine-learning reporting guidelines.

“we incorporated the use of interpretability tools (eg, SHAP 28 values, feature importance) to ensure transparency, a priority emphasized in recent ML reporting guidelines.”

• cites [28] Fryer, 2021, Shapley values for feature selection: the good, the bad, and the axioms, IEEE Access sourceSupported

May combine two claims.

Some studies (42,61,84,88) were unclear about their analysis because they used uninterpretable features or did not use methods like SHAP to suggest interpretations, and they limited their results to classification performance.

“Some 42,61,84,88 were unclear about their analysis because they used uninterpretable features or did not use methods like SHAP to suggest interpretations, and they limited their results to classification performance.”

• cites [42] Zhang, 2022 sourceNot checked — no full text of the source
• cites [61] Coppersmith, 2014, Measuring post traumatic stress disorder in Twitter, ICWSM. sourceNot checked — no full text of the source
• cites [84] Kirinde Gamaarachchige, 2019 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

Few studies presented an interesting result in the sense of focusing on symptoms such as intrusions and avoidance.

“few studies presented an interesting result in this sense.40,42,54,59,80,90”

• cites [40] Quillivic, 2024, Interdisciplinary approach to identify language markers for post-traumatic stress disorder using machine learning and deep learning, Sci Rep sourceSupported
• cites [42] Zhang, 2022 sourceNot checked — no full text of the source
• cites [54] Paquet, 2024, Implications of language use in posttraumatic nightmares on psychological symptoms, Psychol Trauma sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

Narrative coherence has been well studied in Formal Thought Disorder research.

“For example, narrative coherence has been well studied in Formal Thought Disorder research.91,92”

• cites [91] Corcoran, 2020, Language as a biomarker for psychosis: a natural language processing approach, Schizophr Res sourceSupported
• cites [92] Andreasen, 1979, Thought, language, and communication disorders. I. Clinical assessment, definition of terms, and evaluation of their reliability, Arch Gen Psychiatry sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S4 (based).

This is the first systematic review investigating the use of language as a marker for PTSD.

“This is the first in the field investigating the use of language as a marker for PTSD.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Other research suggested that shorter narratives are linked with more severe PTSD.

“Other research has suggested the oppositeshorter narratives are linked with more severe PTSD.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The search resulted in the identification of 593 articles.

“This resulted in the identification of 593 articles”

Probably not a substantial claim: may not pass S3 (load-bearing).

Longer trauma narratives were associated with better psychological adjustment.

“Beaudreau 55 observed that longer trauma narratives were associated with better psychological adjustment”

• cites [55] Beaudreau, 2007, Are trauma narratives unique and do they predict psychological adjustment?, J Trauma Stress sourceNot checked — no full text of the source

May combine two claims.

For the overall analysis of 14 studies, Egger's test and the homogeneity test both showed significant results with P-values below .01, indicating significant publication bias and substantial heterogeneity.

“For the overall analysis of 14 studies, Egger's test and the homogeneity test both showed significant results, with P-values below the .01threshold, indicating significant publication bias and substantial heterogeneity.”

May combine two claims.

The Galbraith plot displays heterogeneity between studies, with four points falling outside the 95% CI lines.

“The Galbraith plot (Figure S6) further displays heterogeneity between the studies, with 4 points falling outside the 95% CI lines.”

May combine two claims.

For the analysis excluding social media data (8 studies), both Egger's test and the homogeneity test showed P-values below .01, confirming significant publication bias and persistent heterogeneity.

“Similarly, for the analysis excluding social media data (8 studies), both tests also showed P-values below the .01threshold, confirming the presence of significant publication bias and persistent heterogeneity.”

May combine two claims.

The main bias for outcome occurred in studies using self-reported diagnoses, with no control for time since traumatic event, source of trauma, and quality of diagnosis.

“For the outcome (ie, PTSD diagnosis), the main bias occurred in studies using self-reported diagnoses, with no control for time since traumatic event, source of trauma, and no control for the quality of the diagnosis.”

May combine two claims.

The overall pooled AUC for the 14 studies was 0.81 (95% CI 0.75, 0.85), indicating high performance of classification models used to detect PTSD.

“The overall pooled AUC for these studies was 0.81 (95% confidence interval [CI]: 0.75, 0.85), indicating high performances of the classification models used to detect PTSD (Figure 4).”

Probably not a substantial claim: may not pass S3 (load-bearing).

A total of 8 studies were assigned a low risk of bias rating.

“A total of 8 studies were assigned a low risk of bias rating.”

Probably not a substantial claim: may not pass S3 (load-bearing).

There has been a notable increase in publications involving computational methods since 2012.

“demonstrating a notable increase in publications involving computational methods since 2012.”

May combine two claims.

Most studies had an EPV of less than 5, classified as uncertain, and studies with an EPV of less than 3 were classified as high risk.

“Most studies had an EPV of less than 5, which was classified as uncertain, and studies with an EPV of less than 3 were classified as high risk.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Larger sample sizes often relied on self-reported data from social media.

“with larger sample sizes often relying on selfreported data from social media.”

May combine two claims.

Both PTSD prevalence (coefficient = -0.0027, P = .005) and sample size (coefficient = -1.501e-05, P = .023) negatively correlate with AUC.

“The results indicate that both PTSD prevalence (Coefficient ¼ -0.0027, P ¼ .005)and sample size (Coefficient ¼ -1.501e-05, P¼ .023)negatively correlate with AUC”

May combine two claims.

Analysis-related biases were common, with low EPV ratios and issues like not reporting statistical power.

“Analysis-related biases were common, with low EPV ratios and issues like not reporting statistical power.”

May combine two claims.

Self-reported PTSD diagnoses varied widely and often lacked quality control.

“Self-reported PTSD diagnoses varied widely and often lacked quality control.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The average PTSD prevalence across studies was 47%.

“and the average PTSD prevalence across studies being 47%.”

May combine two claims.

Common issues included inadequate control groups and insufficient control over participants' mental health conditions.

“Common issues included inadequate control groups and insufficient control over participants' mental health conditions.”

May combine two claims.

Most studies employed a combination of descriptive and inferential statistics, with some utilizing ML methods to build and validate models, and a few employed deep learning models.

“Most studies employed a combination of descriptive and inferential statistics, with some utilizing ML methods to build and validate models, and a few employed deep learning models.”

May combine two claims.

Many studies used LIWC without reporting coverage or evaluating its performance on specific corpora, with some applying it to translated data or very small samples, raising bias concerns.

“Many studies used LIWC without reporting coverage or evaluating its performance on specific corpora, with some applying it to translated data or very small samples, raising bias concerns.”

May combine two claims.

Speech disorganization and narrative coherence were well-documented but lack standardized measures, making comparison of results difficult.

“Speech disorganization and narrative coherence were well-documented but lack standardized measures, making the comparison of results difficult.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Self-administered tools were the most common diagnostic tool.

“Diagnosis tools varied, with self-administered tools being the most common”

Probably not a substantial claim: may not pass S3 (load-bearing).

A protocol for this study was published with PROSPERO under registration number CRD42024501094.

“A protocol for this study was published with the International Prospective Register of Systematic Reviews (PROS-PERO), under registration number CRD42024501094”

Probably not a substantial claim: may not pass S3 (load-bearing).

Three review authors independently reviewed the titles and abstracts of the studies identified by the search strategy.

“Three review authors (R.Q., S.M., F.G.) independently reviewed the titles and abstracts of the studies identified by our search strategy.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Four reviewers independently assessed risk of bias in included studies using PROBAST.

“Four reviewers (R.Q., S.M., Y.A., and F.G.) independently assessed the risk of bias in the included studies using the Prediction Model Risk of Bias Assessment Tool (PROBAST).24”

• cites [24] Wolff, 2019, PROBAST: a tool to assess the risk of bias and applicability of prediction model studies, Ann Intern Med sourceNot checked — no full text of the source

May combine two claims.

Conflicting results on narrative length and cognitive processing words highlight the complexity of PTSD, whose symptoms and expression vary widely depending on type of exposure, time to traumatic event, and population characteristics.

“These conflicting results (narrative length and cognitive processing words) highlight the complexity of PTSD which symptoms and expression vary widely depending on type of exposure, time to traumatic event, and population characteristics.”

May combine two claims.

PTSD memory is fragmented, and this fragmentation should be reflected in the narrative.

“many research indeed considered that PTSD memory is fragmented 38,78 (see also criterion D1 of DSM-5: “Inability to recall key features of the trauma”) and this fragmentation should be reflected in the narrative.”

• cites [38] Gray, 2001, Complexity of trauma narratives as an index of fragmented memory in PTSD: a critical analysis, Appl Cogn Psychol sourceNot checked — no full text of the source
• cites [78] Miragoli, 2017, Narrative fragmentation in child sexual abuse: the role of age and post-traumatic stress disorder, Child Abuse Negl sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Shorter narratives might reflect avoidance and a struggle to organize thoughts.

“Bisson Desrochers et al 75 noted that shorter narratives might reflect avoidance and a struggle to organize thoughts”

• cites [75] Bisson Desrochers, 2016, Gender and changes in trauma narrative following CBT for PTSD, J Aggress Maltreatment Trauma sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

Machine-learning studies often require evaluation beyond traditional metrics, which may better capture class imbalance or model generalizability.

“ML studies often require evaluation beyond traditional metrics (eg, AUC-ROC, F1-score), which may better capture class imbalance or model generalizability.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Narrative coherence generally suggests a structured and logical retelling of events, which is often associated with better psychological adjustment.

“For this reason, narrative coherence generally suggests a structured and logical retelling of events, which is often associated with better psychological adjustment.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Techniques like cross-validation, regularization, and independent test sets are critical in machine learning due to algorithmic complexity and data-driven optimization risks.

“Techniques like cross-validation, regularization, and independent test sets are critical in ML due to algorithmic complexity and data-driven optimization risks.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based). May combine two claims.

Reproducibility is a cornerstone of machine-learning research, and public code sharing is increasingly mandated to address replication crises in computational science.

“Reproducibility is a cornerstone of ML research, and public code sharing is increasingly mandated to address replication crises in computational science.”

May combine two claims.

Larger datasets and higher PTSD prevalence are associated with reduced classification performance.

“suggesting that larger datasets (more heterogeneous) and higher PTSD prevalence (more balanced dataset) are associated with reduced classification performance.”

Classified as probably not a substantial claim. May combine two claims.

Studies using noninterpretative models are deemed less applicable due to limited clinical utility, as model transparency is an essential criterion for clinical use.

“For example, studies using noninterpretative models are deemed less applicable due to limited clinical utility, as model transparency is an essential criterion for clinical use.25”

• cites [25] Chandler, 2020, Using machine learning in psychiatry: the need to establish a framework that nurtures trustworthiness, Schizophr Bull sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Integrating linguistic biomarkers into clinical assessments will enable a more nuanced understanding of PTSD symptoms and their severity, particularly by helping to reduce false negatives through the detection of uncontrollable measures such as avoidance symptoms.

“This integration will enable a more nuanced understanding of PTSD symptoms and their severity, particularly by helping to reduce false negatives through the detection of uncontrollable measures such as avoidance symptoms.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

An overemphasis on diagnosis rather than symptoms overlooked the evolving nature of PTSD definitions.

“an overemphasis on diagnosis rather than symptoms overlooked the evolving nature of PTSD definitions”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Language features can be used to monitor patients' progress and predict treatment outcomes.

“Similarly, language features can be used to monitor patients' progress and predict treatment outcomes.”

May combine two claims.

High scores on noisy data like psychopathological texts suggest over-fitting and/or bias in participant selection.

“Such high scores on noisy data like psychopathological texts suggest over-fitting and/or bias in participant selection.”

May combine two claims.

The use of linguistic biomarkers can significantly improve the accuracy and speed of PTSD diagnoses.

“The use of linguistic biomarkers can significantly improve the accuracy and speed of PTSD diagnoses.”

May combine two claims.

The field is confronted with significant obstacles, including methodological inconsistencies, the lack of validation for existing tools, and biases.

“the field is confronted with significant obstacles, including methodological inconsistencies, the lack of validation for existing tools, and biases.”

Probably not a substantial claim: may not pass S2 (checkable).

The incorporation of linguistic biomarkers into clinical practice has the potential to facilitate more accurate, timely, and nuanced PTSD diagnoses, which will ultimately benefit both patients and clinicians.

“Overall, the incorporation of linguistic biomarkers into clinical practice has the potential to facilitate more accurate, timely, and nuanced PTSD diagnoses, which will ultimately benefit both patients and clinicians.”