External Validation by Machine Learning and Reduction of the Input Dimensions of the D-PSY Scale for Dimensional Psychopathology

Validation Externe Par Apprentissage Statistique et réduction Des Dimensions d'entrée de l'échelle D-PSY

There have been debates about whether psychiatric disorders should be classified according to categories (categorical model) or according to a continuum between normal and pathological functioning (dimensional model). While the former is the main one used to facilitate reliability, it has several limitations. We will use machine learning to develop a predictive model, bridging the gap between the dimensional and categorical approaches. Psychometric measures obtained from a questionnaire will be collected. Then, a supervised descriptor selection approach will be applied to predict categorical outcomes from dimensional inputs. The resulting prediction will be based on nonlinear modeling based on universal approximators. We developed this input questionnaire with four main objectives: 1) to briefly scan most of the categories generally described in the international nosography; 2) to use a continuous scale following a dimensional approach; 3) to use positively oriented sentences to decrease social desirability bias; 4) to be less confronting for the patient.

The questionnaire is built on dimensions, whereas psychiatric diagnoses are built on categories. We will model the transition from one to the other. This approach will allow us to verify the hypotheses of diagnostic categories construction in psychopathology, integrated in the RDoC approach; and later on to standardize the psychometric measures used in cognitive-behavioral therapy treatments.

200 adults will undergo a standard clinical interview (SCID-5, First et al., 2017), and the psychological screening self-report questionnaire (D-Psy) and additional questionnaires measuring social desirability dissociation, phobia and autism.

This is a pilot study, in which 200 participants, aged 18-55 years, taken from the general population, with suspected psychopathology, will participate in screening consultations in which symptoms will be assessed with three different psychometric diagnostic tools. Their symptoms will be assessed using the French version of the SCID-5 (First et al., 2017), and the D-PSY screening procedure, including a self-reported questionnaire and semiological observations. Supplemental questionnaires will also be administered. The order in which these diagnostic tools will be administered will be randomized. Participants will be recruited in France, where the research will take place.

Doctors in the vicinity of the institute will be contacted and the study will be explained to them so that they can suggest eligible patients to participate. Advertising like posters presenting the study will also be put up in health care center and specialized websites. When a person agrees to participate in the study, they will be given an information letter about the protocol. After a delay of reflection, patients who agree to perform this research will be convened for an inclusion visit. After checking the inclusion/exclusion criteria, the letter of information will be completed and signed by the participant and duly countersigned by the investigator.

During this inclusion visit, a history will be taken in which demographic, psychiatric/medical data will be collected. Following this interview, the trained practitioner will complete his or her semiological observations of the patient.

The assessment phase includes the SCID-5 (First et al., 2017), the D-PSY questionnaire and standardized questionnaires. The D-PSY questionnaire and the SCID-5 (First et al., 2017) will be counterbalanced in terms of the order of administration, with half of the participants taking the SCID-5 (First et al., 2017) first and then the D-PSY questionnaire and the others standardized; the other half will take the D-PSY first and then the SCID-5 (First et al., 2017) and the other questionnaires.

The SCID-5 interview will be conduct by a practitioner familiar with this interview and with the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5, American Psychiatric Association, 2013) classification and diagnostic criteria. Mental health practitioners administering the interviews will be blinded to the questionnaire scores.

The D-PSY questionnaire will be completed at the institute. A professional will explain how to complete the questionnaire if necessary and remain available in case of technical matters.

The other standardised questionnaires will be randomised for order of administration. These standardised questionnaires measure social desirability (Balanced Inventory of Desirable Responding, BIDR, Paulhus, 1985, 1991; in French: Cournoyer & Sabourin, 1991), dissociation (DIS-Q, Vanderlinden et al., 1993), phobia (the Fear Questionnaire, F.Q., Marks & Mathews, 1979; in French Cottraux et al., 1987) and autism (AQ, Baron-Cohen et al., 2001; in French, translation by Braun & Kempenaers, 2007)

This screening questionnaire has been created in order to measure quickly the most common mental health disorders and for the assessment of personality disorders. This new tool has been developed by the research team through a careful analysis of DSM-5 (American Psychiatric Association, 2013, Crocq et al., 2015) and CIM-11 (Mokaddem, et al., 2020) criteria for diagnosing mental health disorders, as well as criteria described in the SCID-5 (First et al., 2017) and the M.I.N.I (Sheehan et al., 1998). These are used respectively for the assessment of the 17 most common mental health disorders and for the assessment of personality disorders (as defined in the DSM-V, American Psychiatric Association, 2013).

After analysing the diagnostic and transdiagnostic criteria of the pathologies, the questions were carried out by focusing on one symptom and with the desire to create a continuum between two extremes of the symptom.

For example, the question "To what extent do you involve your relatives in your important decisions?" refers to autonomy and confidence in one's decisions. The extremes of the visual analogue scale are:

1. "I am confident in my ability to make good decisions"
2. "The opinion of my relatives is important to me". If a participant puts a response closer to the extreme (a), this will refer more to a narcissistic personality; whereas if he puts a response closer to the extreme (b), this will refer to a dependent personality.

The questions were developed in this way for all the pathologies present in the M.I.N.I. (Sheehan et al., 1998) and the SCID-5 (First et al., 2017) and some absent from it have been added: autism and specific phobia. Additional questions concerning transdiagnostic criteria were added, such as questions on dissociation and sleep.

The scale was constructed with the objective of reducing social desirability bias and confrontational questions. For this reason, we chose to use only positive terms or phrases that can be seen as positive or socially acceptable in the answers to the questions.

The semiological observations concern the elements identified in the speech of the participant and the observations that the practitioner makes during the interview. In the speech, the practitioner identifies whether the participant addresses somatic disorders (digestive, sensory, cardiac disorders, etc.), if he has psychic disorders (leakage of ideas, dissociation, delusions, etc.), if he addresses elements concerning his sleep (nocturnal awakenings, drowsiness, nightmares, etc.), elements concerning functional impairment (relationship difficulties, stress, etc.), whether he has mood disorders (dysphoria, euphoria, suicidal ideation) and if it addresses life events (positive such as encounters or opportunities or adverse such as bereavement, failure, accident). In terms of observations, the practitioner notes whether there is psychomotor impairment (agitation, inhibition, tics, etc.), a deterioration in dialogue (logorrhea, autistic withdrawal, etc.) and a deterioration in the relationship during the interview (provocation, resistance, seduction, etc.).

The semiological observations will be carried out by the practitioner during the inclusion visit. The latter will have to note the different clinical characteristics that he observes.

The SCID-5 (First, 2014) is a semi-structured interview in which the major DSM-5 diagnoses are assessed. This interview lasts approximately 45 to 90 minutes and will consist of a single interview.

This interview will be conducted by an experienced mental health practitioner familiar with the DSM-5 (American Psychiatric Association, 2013) classification and diagnostic criteria.

Each practitioner will be trained for the delivery of the SCID-5 (First et al., 2017) via the case studies and role plays included in the user guide provided by the authors.

Following this primary assessment phase including SCID 5 (First et al., 2017), the D-PSY questionnaire and the assessment of the semiological observations; other tests will be presented to the participants.

( a ) The French version of the Balanced Inventory of Desirable Responding (BIDR, Paulhus, 1985, 1991; in French: Cournoyer & Sabourin, 1991) is 40-item self-report measuring deliberate and nondelibarate socially desirable response style.

( b ) The dissociation questionnaire (DIS-Q, Vanderlinden et al., 1993) is a self-report instrument of dissociation symptoms. Four subscales exist: 1) identity confusion; 2) loss of control over behaviour, thoughts and emotion; 3) amnesia; and 4) absorption. We used the French version (Mihaescu et al., 1998).

( c ) The French version of the Brief standard self-rating for phobic patient or "the fear questionnaire" (F.Q., Marks & Mathews, 1979; in French Cottraux et al, 1987) is a 24-item self-report composed of three parts. The first part measure global phobia including agoraphobia, blood and injury phobia, and social phobia; the second part measure anxiety-depression and the third one the consequences of this behaviour. It yields four scores: main phobia, global phobia, total phobia and anxiety-depression.

( d )The Autism Spectrum quotient (AQ ; Baron-Cohen et al., 2001; in French, translation by Braun & Kempenaers, 2007) is a diagnostic self-report measuring the expression of Autism-Spectrum traits.

Statistical learning will be used to build a predictive model, separating the data to build an independent test base. The main objective of this investigation is to predict the majority of SCID-5 pathologies (validity test) with a success rate significantly higher than chance.

The secondary objectives of this investigation are:

• To extract a dimensional structure of pathologies
• To resolve the social desirability bias present in current questionnaires by using positive terms.
• To make the questionnaire easy to use (speed, feasibility, constraints)
• To set up an emotionally non-confronting questionnaire

Inclusion Criteria:

• Adults over 18 years old and less than 65 years old who are fluent in French (oral and written), with suspicions of psychopathology.
• Protected adults and pregnant women cannot participate.

Exclusion Criteria:

• Participant with neurological or organic mental disorder (not induced by substances) including dementia, organic amnesic syndrome, delirium;
• other mental, personality or behavioural disorders due to brain damage, dysfunction or to physical disease.