We conducted a qualitative and quantitative study via a computerized survey (Google-Form), in collaboration with the Sociology and Anthropology Laboratory of the University of Burgundy Franche-Comté (LaSA, UBFC).

This study focused on a population of psychiatrists working in France. They ranged from residents to senior psychiatrists, working in psychiatric facilities, general or university hospitals, or private practices. Requirement of Ethical Committee’s approval was waived.

We used an original form of assessment, with 2 researchers at Sociology and Anthropology Laboratory of the University of Burgundy Franche-Comté, based on the screenplay method, an assessment method used to expose respondents to challenging and problematic clinical cases, in order to ask them to express what should be done or what they themselves have done to act with competency in such situations. By confronting the psychiatrists with systems or devices that are still essentially restricted to the field of research, we were able to review some aspects of reality that are not captured by other types of evocation. The 3 scenarios used here allow practitioners to think about devices that are currently in the research domain and are not used (or little used) in daily routine.

The screenplay method featured 3 clinical case presentations involving new technologies (Table 1 and Multimedia Appendix 1).

All the questions were designed during 3 focus groups including psychiatrists and sociologists and were tested with cross-validation on a sample of psychiatrists working at Saint-Antoine Hospital in Paris, France. The first part of the survey (15 questions) collected epidemiological data: sex, job, place of practice, theoretical and practical training (neurobiological, psychoanalytic, integrative, cognitive behavioral therapy, etc), workplace, year of graduation, and practice area (adult psychiatry, child psychiatry, forensic psychiatry, etc). The second part assessed the acceptability of the support systems and the psychiatrists’ professional culture, with 15 questions per scenario (total of 60 questions). To avoid responder focus being too much on confounding factors (ie, the shortage of health personnel aspect), we asked direct questions about the devices in our questionnaire (Textbox 1). A blank field allowed us to collect qualitative data in the form of feedback at the end of the survey.

The various technologies described above can be studied from a sociological perspective by examining the factors that prevent or, conversely, encourage their use. Several dimensions that can influence acceptability were included in an acceptability model specifically developed for the study and inspired by research on human-machine interaction and management information systems, combining the Nielsen, International Organization for Standardization, and Shackel models [21,22] (see Multimedia Appendix 2). The variables most frequently associated with acceptability are usability (ie, intention to use), supposed usefulness, and reliability [23-25]. In our model, we assessed 4 variables: usefulness, usability, reliability, and risk (Textbox 1). For each variable, participants responded to the questions on a Likert-like scale ranging from 1 to 6, depending on the item. To gauge the acceptability of each system, we calculated a composite score with 3 values (positive, intermediate, and negative).

The purpose of each item was to bring out the characteristics of the sociological concept known as psychiatrists’ professional culture: what made them psychiatrists, with which technologies they would refuse to compromise, and how they saw themselves in relation to other specialists. We took 2 major areas of professional culture into account. First, we investigated the psychiatrists’ scientificity level, reflected by the use of biometric data, MRI, blood tests, and physical examinations (assuming that the more scientifically-minded the psychiatrists are, the more willing they are to use complementary examinations). Second, we probed the psychiatrists’ specific relationship with technology by analyzing the hopes and fears generated by these new tools (ie, did they think that these technologies would help them or replace them?; see Textbox 2).

The survey was created with Google Forms and sent by email to the relevant professionals via several mailing lists (residents’ association, private practice associations, clinical facilities, personal social networks, etc). Respondents could answer via an internet browser. After a short introductory text, the scenarios appeared one after the other, each followed by the corresponding questions. The survey was anonymous and took about 10-15 minutes to complete.

We performed an initial descriptive analysis of the population using multiple regression analysis. Comparisons of proportions were carried out using a z test with Bonferroni correction. Pearson correlation coefficients were used to analyze correlations between variables. The variables were compared with nonparametric chi-square tests or with Fisher’s test when the conditions for chi-square application were not met, using Microsoft Excel, SPSS v24, and R statistical software. The significance level was set at 5%, such that differences with a P value <.05 were deemed to be significant. In order to achieve 95% statistical power with an alpha risk of .05, the bibliographic analysis indicated that 374 participants were required (bearing in mind that there were 12,591 psychiatrists in France in 2016) [26]. Qualitative variables were partially analyzed by LaSA using Modalisa (in press).

The Web-based survey was available between June 30 and August 8, 2016. A total of 528 responses were received. We excluded 5 empty surveys, 5 duplicates, and 3 incomplete surveys (no responses to at least 1 whole scenario) such that 515 surveys were included in the analysis.

The study population was predominantly female (299/515, 58.1%), mainly composed of young psychiatrists who had already graduated or were set to do so between 2010 and 2020 (342/515, 66.4%), and the majority of practitioners worked in adult psychiatry (270/515, 52.4%). Residents made up a large proportion of the sample (241/515, 46.8%), followed by hospital practitioners (148/515, 28.7%) and private practitioners (49/515, 9.5%). The 2 most common theoretical approaches were “several approaches focusing on neurobiology or cognitive behavioral therapy” and “integrative practice” (see Multimedia Appendix 3).

We asked 3 questions that allowed us to study the place of psychiatrists compared with other medical specialists in terms of scientificity by probing the importance they gave to physical signs (ie, biometric data), technology, and algorithmic thinking. These questions made it possible to distinguish between psychiatrists who:

They also allowed us to highlight the degree of opposition to these dimensions, including affirming the primacy of the human relationship and the refusal or rejection of technology. We were able to place psychiatrists on a continuum running from a medical to psychological subtype and conducted an analysis in which these profiles were crossed with the epidemiological data. This analysis revealed a perfect match between acceptability scores and psychiatrists’ professional culture: the medical profile had the highest acceptability score.

We achieved our primary objective of determining the acceptability of new technologies by psychiatrists, and to our knowledge, this is the first study to assess psychiatrists’ views on computerized CDSSs. CDSSs clearly had different acceptability profiles. The connected wristband seemed to have the lowest acceptability profile; CAT and EMA smartphone- based assessment were rated as the most reliable but with the least usability; and ML-based MRI or blood tests were rated as having the greatest usefulness and usability but with the greatest risk. Furthermore, approximately half of the psychiatrists claimed to wait for (and expected to receive) more scientifically validated arguments before reaching any conclusions. This could be a valid argument as the wait-and-see position could be underlying a healthy skepticism. In our opinion, what appears above all in this matter is the lack of knowledge of these techniques among practitioners. Many scientific studies clearly demonstrate the interest in these devices. A recent review of the literature by Faurholt-Jepsen et al [27] regarding the comparative validity of electronic self-assessment techniques over scale or classical clinic assessment found no significant differences between all evaluation methods (smartphone, computer software, internet, and standard clinical scales). According to a review conducted by a team from the Black Dog Institute between 2008 and 2013 [28], reliability and patient acceptability of smartphone-based symptom monitoring is high, with good agreement between electronic measurements and standard diagnostic scales. Similarly, Torous et al [17] demonstrated a good agreement between paper and smartphone versions of the Patient Health Questionnaire, 9^th Revision, Beck Depression Inventory, and Quick Inventory of Depressive Symptomatology, with a preference for the smartphone media. Regarding the digital phenotype, the capture of passive data concerns objective criteria of depression recognized by all practitioners (ie, psychomotor retardation in depressive disorder, manic graphorrhea, etc), so it is more a doubt cast on the ability of these devices to “capture” these elements than a rejection of the principle behind it. However, again, several studies scientifically show the validity of this type of data collection. It should also be noted that all the scripts had bibliographic references at the end, allowing responders to learn about these studies.

The subgroup analysis showed that acceptability was strongly mediated by a psychiatrist’s profile and, more specifically, by sex, theoretical approach, mode of practice, role, and experience. One may wonder why the youngest (residents) have a more negative acceptability profile than the others, while intuitively we could think that they would be more favorable to new technologies given their different usage profile. The qualitative sociological analysis of the data shows that it could be a counter-reaction in relation to the feeling of being dispossessed of a recently hard-won knowledge (ie, “If a machine is able to do better than me, then why did I spend 4 years in training?”). This effect diminishes with age.

The questions probing the psychiatrists’ professional culture indicated that a certain type of professional culture corresponded to a certain acceptability profile. Our analysis suggests that there were several subgroups, which differed in almost every aspect, with psychiatrists whose professional culture could be defined as medical at one extreme and those who had a more psychodynamic culture at the other extreme. The question of prediction [29] seems more sensitive for certain categories, especially child psychiatrists and forensics practitioners who are the most reluctant in this area, which illustrates the important current debate on neuroprediction [30,31].

The number of usable responses we collected (N=515) allowed us to have a representative sample of a good size and a geographical distribution that did not influence the data (no significant differences between the regions). Compared with other survey-based studies, this was a large sample as most studies on psychiatrists collect between 50 and 150 responses [32]. The questions were developed in collaboration with sociologists to ensure the relevance of the data we collected and allow for the construction of a sociological hypothesis. Our analysis allowed us to identify a typology of French psychiatrists that featured 2 contrasting schools of thought. The use of quantitative data, including ratings on 3- or 6-point scales, allowed us to undertake a relatively fine-tuned analysis. Furthermore, the questions were developed from a model specifically adapted to medical technology acceptance and inspired by several valid theoretical models [21,22].

Qualitative analysis allows us to explore the different fields of acceptability, raising several constraints. The main obstacle was the psychiatrists’ fear that they would do more harm than good either by generating anxious counter-reactions (especially with regard to the EMA smartphone app and connected wristband) or by creating a risk of overtreatment by diagnosing problems that did not exist. This evokes the notion of self-fulfilling prophecies developed by Robert K Merton [33], which consists, from a false starting hypothesis, of provoking a behavior that makes this initially false hypothesis become true. An anxious reaction may be one of the consequences, which may increase the risk of recurrence, but we could argue that this could also apply to any form of care (including medication). Much of the feedback focused on the third scenario (which had the highest acceptability profile), with questions about which course of action to pursue if the MRI or the blood tests predicted a transition to psychosis and pointing out the risk of jumping to diagnostic conclusions. Several respondents indicated that they would refuse to introduce a preemptive antipsychotic treatment based on a prediction made in this way. Several commented that there was no point in predicting an incurable disease. In fact, this is an interesting result itself, as it shows that the vast majority of psychiatrists interviewed have no idea what to do if there is a risk of psychotic transition. Recent works highlight the importance of a number of measures to limit this risk (fatty acids, low-doses of atypical antipsychotic, active surveillance, open dialogue, etc), and clearly, many psychiatrists are not aware of this [34]. Feedback on the second scenario raised the same questions, with the idea that it is not ethically acceptable for a psychiatrist to delegate a monitoring task to a machine. This technology elicited particular ethical and political considerations based on the notion that these connected wristbands are associated in people’s minds with the electronic ankle tagging of prisoners and that they are part of a political agenda used to compensate for increasing shortages of health professionals in some areas. From our point of view, although there is indeed an incentive for the development of telemedicine given the current medical demography in France, when we evaluate the impact of a depressive or psychotic recurrence on the life of a patient, it is clear that technologies that allow better monitoring of their condition should not cause a rejection but rather an interest, especially since these technologies have been developed by doctors (and in a number of cases with patients) and not by politicians. Regarding the first scenario, some respondents argued that this technology could result in the loss of opportunity, especially if it prevented practitioners from diagnosing other problems because the focus was on the system’s diagnosis, and some mentioned the risk of practitioners losing their clinical sensitivity through lack of practice. This type of reasoning applied to other fields of medicine when additional tests or new investigation tools were developed. We do not think that the development of the stethoscope made cardiologists lose their clinical sense, we think that it improved them. These technologies must become complementary psychiatric examinations and complement the currently subjective approach employed by psychiatrists in diagnosis or prediction.

Significant disparities between devices highlight varying degrees of acceptability, both technology-dependent (previously known technology such as MRI seems better accepted, while connected objects are less well-known) and the underlying theoretical presupposition. Thus, the prediction of a psychotic transition remains subject to many fears, whereas the computerized questionnaires are not. Paradoxically, it is the MRI coupled with ML that psychiatrists find most useful. There are 2 issues very clearly raised here: the psychiatrists questioned in this survey are very minimally informed, notably on the scientificity of these devices, although many studies have already been published and the practitioners do not necessarily know what to do with the new data that these machines can obtain, that is, they have very little knowledge about what to do if there is a risk of psychotic transition, they are not familiar with the concepts of precision medicine, and so on. In this context, it seems very important to increase training measures on new technologies, particularly by integrating this into the resident teaching program. For the past 3 years in France, most psychiatric congresses have offered an innovation session or a new technology session; this seems to be an interesting way to promote these new tools but needs to be expanded.

Furthermore, our results assume that new technologies are challenging the psychological subtype of psychiatrist while consolidating the medical subtype. Neurobiological psychiatrists are not challenged by these technologies, which they regard rather as tools that extend or complement their practice. Regardless of the system, psychiatrists with a psychoanalytic orientation are clearly reluctant, and we suggest that it is both the use of scales and the technological dimension that account for their negative stance. This assumption is reinforced by the large number of comments that evoked the technology’s impact on the therapeutic relationship. Special measures should be taken to reassure that specific subtype of psychiatrist by demonstrating that these systems are part of a global care that does not negate the psychological aspect of the disease and that it allows an improvement in patient care.

Many ethical issues were raised by this study, and data security, data storage, privacy, and hacking risk have yet to be resolved. Disease detection or risk prediction, whether in the case of depressive relapse or transition to psychosis, is necessarily stressful for patients and brings the risk of excessive focus and an anxious counter-reaction. To offset these risks, it is essential for psychiatrists to be involved in the development of new technologies, to prevent their loss of control over them. Developers have a major interest in communicating better about the design of these tools and the algorithms they want to be used in the future. In order to complete our research, a comparative study using the same methodology is being developed to better understand the acceptability of these technologies by patients, nurses, and general practitioners.

The main limitation could be tautological in that the differences we found between various acceptability profiles may simply have reflected a difference in theoretical approach (neurobiological psychiatrists vs the rest). This suggests that the technologies we studied were based on a theoretical presupposition that was purely neurobiological. While it is true that the third scenario had a clear neurobiological emphasis, the same cannot be said for the other two. In the first scenario, the use of smartphone-based EMA raised the question of scales, but scales presuppose nothing of the etiopathogenesis of the disorder being assessed. Furthermore, CAT does not use scales, as it simulates the psychiatrist’s way of thinking by choosing a specific question from a database made up of more than 500 items. The same applies to the connected wristband as biometric data capture only means that some depression symptoms can be objectified (eg, psychomotor retardation)—a fact that no psychiatrist can refute. It was, therefore, not the opposition between neurobiological and psychodynamic issues that were assessed in our study but the relationship with technology exhibited by different subtypes of psychiatrists.

The type of professional culture (theoretical background and practice) appeared to exert a strong influence on the acceptability of the technologies we studied. Overall acceptability was moderate, and respondents expressed many reservations and raised many ethical and ideological questions. Indeed, a probability derived from the analyzed data cannot systematically be transformed into diagnostic or therapeutic certainty. Some of their concerns are relevant and their skepticism can be understood although many of the issues raised are in fact devoid of reality and reflect a great lack of knowledge of the current state of research on these new technologies. It is surely necessary for psychiatrists to adopt a clear stance with regard to these radical changes that are upsetting traditional practice, and for them to be able to do this, they must be informed, interested, and allowed to contribute to the development of these new technologies [35,36], going as far as joining the “disruptors of health sciences” [37]. The acceptability model we developed, using a complex sociological methodology featuring clinical case scenarios intended to elicit emotional responses, needs to be replicated.