Review Reports - Between Awareness and Readiness: Perceptions of Artificial Intelligence Among Indonesian Mental Health Professionals

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

I highly appreciate your efforts and interest to adress a crucial, modern, and interesting topic. However, the paper has several significant weaknesses and inconsistencies which impacts the quality and readability; All these must be clarified.

Target population and refinement of the narrative scope: Table 2 title reads "Knowledge & AI use in child and adolescent mental health service in Indonesia" and Table 3 title reads "AI use in child and adolescent mental health services in Indonesia". Furthermore, line 178 refers specifically to "child & adolescent mental health services". However, the broader manuscript title, introduction, sample description (Table 1), and conclusion discuss mental health professionals generally across hospitals, clinics, and health centers without limiting the inclusion criteria or analysis strictly to child/adolescent specialists. Please align the scope across the entire paper. Clarify whether this survey targeted child/adolescent specialists specifically or general mental health providers. Remove erroneous references to child and adolescent services in the table titles and text if the study population is general.
Methodological clarification needed : Please clarify how non-probability convenience sampling via email distribution across hospitals affects generalizability and selection bias (e.g., professionals interested in technology may have been more likely to complete an online survey
Statistical values (SHAIP reliability): Lines 109–111 state: "The professional impact of AI with ∝ = 0.710; preparedness for AI with ∝ = 0.610 similar to the original version of SHAIP questionnaire, while professional impact of AI with ∝ = 0.832; preparedness for AI with ∝ = 0.632." This sentence is repetitive and contradictory. Clarify which Cronbach’s alpha values represent the original validation study (Shinners et al.) and which represent the adapted Indonesian version evaluated in this study cohort.
Missing information in multiple linear regression (Table 7): Table 7 presents regression results for factors associated with perception of professional impact.The narrative model states R ²= 33.4% (0.334), but Table 7 lists absolute point estimates/effect sizes such as "1.27" for gender (male) and "1.24" for residents on a 5-point Likert scale average, which seems disproportionately large relative to the unadjusted group means in Table 6 (e.g., Male mean 3.28 vs. Female mean 3.14, a difference of 0.14, not 1.27).
Lack of granularity regarding AI tools and use cases: The study reports that 53.2% of respondents use AI in clinical practice. However, the survey does not report which AI tools (e.g., LLMs/ChatGPT, diagnostic support, administrative transcription, scheduling algorithms) are being used or for what specific clinical workflows.Please d etail what specific tools were captured under "AI use in clinical practice". If detailed sub-questions were not included in the original survey, explicitly discuss this as a key methodological limitation
Inconsistent data reporting between text and results tables:
- Table 3 vs. Narrative (Age group data): In Table 3, under the 41–50 age group, the counts are listed as 10 (20.4%) Yes and 39 (58.4%) No. The sum 10 + 39 = 49, but the row percentage for "No" reads 58.4% instead of 79.6% (39/49).
- Table 3 vs. Table 4 (Profession Categories & Data): Table 3 lists 7 professions (Psychiatrist, GP, Resident, Nurse, Medical Student, Psychologist, Others). Table 4 omits the "Medical Student" row in terms of data formatting/clarity or presents mismatched variable counts.
- Table 6 vs. Table 7: In Table 6 (Univariate Analysis), the association between workplace and professional impact shows p < 0.01. However, Table 6 lists categories: General Hospital, Psychiatry Hospital, Clinic, Puskesmas, Others. The narrative in lines 200–202 states: "Tukey post-hoc analysis revealed significant differences between academic settings and psychiatric hospitals, as well as between academic settings with clinics." However, "academic settings" is not listed as a distinct category in Table 6 or Table 1. Please revise every statistical table against the narrative text for mathematical correctness, consistent categorization, and accurate percentage computations.
- Additionally , there is typographical & grammatical errors in lines 31, 39-40,63, 125, 150,285-286

Author Response

Dear Reviewer,

Thank you for your valuable feedback to improve the quality of this manuscript. We provide here the point-by-point responses to your comments.

Comment 1: Target population and refinement of the narrative scope: Table 2 title reads "Knowledge & AI use in child and adolescent mental health service in Indonesia" and Table 3 title reads "AI use in child and adolescent mental health services in Indonesia". Furthermore, line 178 refers specifically to "child & adolescent mental health services". However, the broader manuscript title, introduction, sample description (Table 1), and conclusion discuss mental health professionals generally across hospitals, clinics, and health centers without limiting the inclusion criteria or analysis strictly to child/adolescent specialists. Please align the scope across the entire paper. Clarify whether this survey targeted child/adolescent specialists specifically or general mental health providers. Remove erroneous references to child and adolescent services in the table titles and text if the study population is general.

Response 1: Thank you for noticing the errors in our manuscript. We have removed references to child and adolescent services from Table 2 and Table 3 as our study population is general.

Comment 2: Methodological clarification needed: Please clarify how non-probability convenience sampling via email distribution across hospitals affects generalizability and selection bias (e.g., professionals interested in technology may have been more likely to complete an online survey

Response 2: We have added this concern as a limitation of the study (Section 4.3)

This study did not examine specific workplace conditions, such as institutional AI infrastructure, staffing levels, or resource allocation, that may explain the observed difference in perceived professional impact and preparedness for AI. Subsequently, sample recruitment via email distribution would introduce selection bias and limits generalizability, with participants more familiar with technology (particularly younger respondents) may have been more likely to participate in the study. We also did not specify which AI tools the respondents had used for their clinical practice, which added to methodological limitation of our study. Specifically, the questionnaire did not distinguish among different types of AI applications, such as generative AI, documentation or administrative tools, educational applications, or diagnostic and clinical decision-support systems. Future research should incorporate these variables to determine whether this difference reflects setting-level resources or other factors in psychiatric practice itself.

Comment 3: Statistical values (SHAIP reliability): Lines 109–111 state: "The professional impact of AI with ∝ = 0.710; preparedness for AI with ∝ = 0.610 similar to the original version of SHAIP questionnaire, while professional impact of AI with ∝ = 0.832; preparedness for AI with ∝ = 0.632." This sentence is repetitive and contradictory. Clarify which Cronbach’s alpha values represent the original validation study (Shinners et al.) and which represent the adapted Indonesian version evaluated in this study cohort.

Response 3: We have modified the sentences to make it clearer which data represent the Indonesian version (Section 2.3), we hope the new sentence is now more comprehensible.

The Indonesian adaptation of SHAIP has been validated and had good reliability on both dimensions. The Indonesian version had α = 0.710 for professional impact of AI and α = 0.610 for preparedness for AI dimension, comparable to the original version of SHAIP questionnaire (α = 0.832 for professional impact of AI and α = 0.632 for preparedness for AI).

Comment 4: Missing information in multiple linear regression (Table 7): Table 7 presents regression results for factors associated with perception of professional impact.The narrative model states R ²= 33.4% (0.334), but Table 7 lists absolute point estimates/effect sizes such as "1.27" for gender (male) and "1.24" for residents on a 5-point Likert scale average, which seems disproportionately large relative to the unadjusted group means in Table 6 (e.g., Male mean 3.28 vs. Female mean 3.14, a difference of 0.14, not 1.27).

Response 4: Thank you for the correction. We have revised the analysis by calculating the mean SHAIP Impact score for each participant, rather than using the summed total score. The results of Table 7 have been analyzed using the corrected mean score, and the corresponding text on Section 3.4 has also been updated (highlighted with yellow).

Table 7. Multiple regression analysis of factors associated with perception of professional impact on AI.

Demographic factors

B (Unstandardized coefficient) (95% CI)

Age group (compared to ≤ 40 y.o.)

● 41–50 y.o.

● 51–60 y.o.

● >60 y.o.

-0.03 (-0.23—0.17)

0.01 (-0.28—0.30)

0.10 (-0.67—0.88)

0.78

0.93

0.79

Gender (male)

0.21 (0.06—0.37)

0.01*

Profession (compared to psychiatrist)

● General practitioner

● Resident

● Nurse

● Medical student

● Psychologist

● Others

0.24 (-0.02—0.49)

0.21 (0.02—0.40)

0.10 (-0.14—0.33)

0.24 (-0.31—0.80)

-0.22 (-0.56—0.12)

0.16 (-0.19—0.50)

0.07

0.03*

0.42

0.39

0.21

0.36

Workplace (compared to general hospital)

● Psychiatry hospital

● Clinic

● Community health center (Puskesmas)

● Others

-0.22 (-0.40 to -0.04)

-0.08 (-0.32—0.15)

0.09 (-0.32—0.50)

0.43 (-0.02—0.88)

0.02*

0.50

0.67

0.06

Comment 5: Lack of granularity regarding AI tools and use cases: The study reports that 53.2% of respondents use AI in clinical practice. However, the survey does not report which AI tools (e.g., LLMs/ChatGPT, diagnostic support, administrative transcription, scheduling algorithms) are being used or for what specific clinical workflows. Please detail what specific tools were captured under "AI use in clinical practice". If detailed sub-questions were not included in the original survey, explicitly discuss this as a key methodological limitation

Response 5: We have added this point as a limitation of our study (described on Section 4.3, P12, L357-362)

We also did not specify which AI tools the respondents had used for their clinical practice, which added to methodological limitation of our study. Specifically, the questionnaire did not distinguish among different types of AI applications, such as generative AI, documentation or administrative tools, educational applications, or diagnostic and clinical decision-support systems. Future research should incorporate these variables to determine whether this difference reflects setting-level resources or other factors in psychiatric practice itself.

Comment 6: Inconsistent data reporting between text and results tables:

Table 3 vs. Narrative (Age group data):In Table 3, under the 41–50 age group, the counts are listed as 10 (20.4%) Yes and 39 (58.4%) No. The sum 10 + 39 = 49, but the row percentage for "No" reads 58.4% instead of 79.6% (39/49).

Response 6: Thank you for your correction and apology for the miscalculation. We have corrected this error on Table 3

Age group

● ≤ 40 y.o.*

● 41–50 y.o.**

● 51–60 y.o.

● >60 y.o.

125 (63.1%)

10 (20.4%)

6 (33.3%)

1 (50.0%)

73 (36.9%)

39 (79.6%)

12 (66.7%)

1 (50.0%)

Comment 7: Inconsistent data reporting between text and results tables:

Table 3 vs. Table 4 (Profession Categories & Data):Table 3 lists 7 professions (Psychiatrist, GP, Resident, Nurse, Medical Student, Psychologist, Others). Table 4 omits the "Medical Student" row in terms of data formatting/clarity or presents mismatched variable counts.

Response 7: We have checked that both Table 3 and Table 4 include “Medical Student” row.

Comment 8: Inconsistent data reporting between text and results tables:

Table 6 vs. Table 7:In Table 6 (Univariate Analysis), the association between workplace and professional impact shows p < 0.01. However, Table 6 lists categories: General Hospital, Psychiatry Hospital, Clinic, Puskesmas, Others. The narrative in lines 200–202 states: "Tukey post-hoc analysis revealed significant differences between academic settings and psychiatric hospitals, as well as between academic settings with clinics." However, "academic settings" is not listed as a distinct category in Table 6 or Table 1. Please revise every statistical table against the narrative text for mathematical correctness, consistent categorization, and accurate percentage computations.

Response 8: I appreciate your revision to enhance the clarity of the sentence. The categories "Academic Settings" in the text have been replaced with "Other workplaces (university/school)" in all statements (P9, L233-234).

Tukey post-hoc analysis revealed significant differences between psychiatric hospitals and other workplaces (university/school), as well as between clinics with other workplaces (university/school).

Comment 9: Additionally, there is typographical & grammatical errors in lines 31, 39-40,63, 125, 150,285-286

Response 9: Thank you for the correction, the sentences have been revised to eliminate the typographical and grammatical errors and we have highlighted with yellow (now on line 37, line 42, line 45, line 64, line 335 and line 337)

Reviewer 2 Report

Comments and Suggestions for Authors

Recommendation: Major Revision

General Evaluation

This study provides useful preliminary evidence regarding awareness, use, and perceptions of artificial intelligence (AI) among mental healthcare professionals in Indonesia. Given the rapid development of digital health technologies, examining these perceptions in the Indonesian healthcare context is timely and potentially valuable. However, the manuscript contains important inconsistencies regarding the definition of the study population, the interpretation of the SHAIP items, and the reporting of numerical and statistical results. These issues should be addressed before the validity of the conclusions can be adequately evaluated.

Major Comments

1. Inconsistency in the Definition of the Study Population

Location: Title (P1, L2–3), Abstract (P1, L14–16), and Methods (P3, L81–86) versus Table 2 (P5, L155), Table 3 (P6, L169), and Results (P7, L179–180).
Issue: The title, abstract, and eligibility criteria broadly define the target population as Indonesian mental health professionals. However, Tables 2 and 3 and part of the Results section refer specifically to “child and adolescent mental health services.” The Methods do not indicate that participants were required to work in child and adolescent mental health services.
Requirement: Please clarify whether the study examined the general mental health workforce or specifically professionals working in child and adolescent mental health services. If the study was limited to the latter group, the title, abstract, eligibility criteria, and interpretation of the findings should be revised accordingly. If it was not limited to this group, references to child and adolescent mental health services should be removed or corrected throughout the manuscript.

In addition, four medical students were included in the sample and subsequent analyses (Table 1, P5; Table 3, P6; and Table 7, P10). Undergraduate medical students would not ordinarily be classified as mental health professionals or healthcare practitioners. Please explain the rationale and eligibility criteria for including medical students. The authors should also consider conducting a sensitivity analysis excluding these participants to confirm that their inclusion does not affect the findings.

2. Misinterpretation of the SHAIP Preparedness Items

Location: Abstract (P1, L22–23), Table 5 (P8, Items 6 and 8), Results (P8, L189–192), and Discussion (P11, L268–271).
Issue: The Discussion states that more than half of the participants believed that they were “personally prepared” for the introduction of AI. However, Item 6 asks whether healthcare professionals overall are prepared for the introduction of AI technology; it does not assess the respondent’s personal preparedness. Item 8 asks whether respondents believe that they have been adequately trained to use AI specific to their role. Only 21.0% agreed with Item 8, whereas 44.2% disagreed and 34.8% were neutral.

The statement in the Abstract that “most reported feeling personally unprepared” is also not directly supported by the questionnaire. The survey assessed formal training history, perceived adequacy of role-specific training, and perceptions of profession-wide preparedness, but it did not appear to directly assess personal readiness as a separate construct.

Requirement: Please distinguish clearly among:
1. receipt of formal AI education or training;
2. perceived adequacy of role-specific AI training;
3. perceived preparedness of healthcare professionals overall; and
4. personal readiness to use AI.

The Abstract, Results, Discussion, and Conclusion should be revised so that each statement accurately reflects the wording and response distribution of the relevant questionnaire item. The findings support limited formal and role-specific training, but they do not directly demonstrate that most participants personally felt unprepared.

3. Numerical and Statistical Inconsistencies

Several numerical and statistical inconsistencies require clarification and correction.

Table 5, Item 10 (P8): The reported numbers total 229 participants (57 + 35 + 137), rather than the full sample of 267. The percentages total only 85.7%. Please clarify whether 38 responses were missing or whether this is a reporting error. If responses were missing, explain why this item had substantially more missing data than the other items and state whether participants were permitted to skip individual questions.

Because Item 10 contributes to the preparedness dimension, the Methods should also explain how these missing responses were handled when calculating the domain score and conducting subsequent analyses. Please state whether complete-case analysis, person-mean substitution, multiple imputation, or another method was used.
Table 3 (P6): For participants aged 41–50 years, 10 answered “Yes” and 39 answered “No.” Given a total of 49 participants, the percentage answering “No” should be 79.6%, rather than the reported 58.4%. Please verify all numbers, percentages, and denominators presented in the tables.
Abstract versus multivariable analysis: The Abstract states that age group, profession, and years of work experience were significantly associated with AI use (P1, L18–19). However, the multivariable model indicates that profession was the only variable with a statistically significant overall association. Age group was not significant overall (p = 0.06), although the 41–50-year category differed significantly from the reference category, and work experience was not significant after adjustment. Please clearly identify which findings are derived from univariable analyses and which are derived from the adjusted multivariable model.

Similarly, the Abstract states that perceived professional impact differed significantly by gender and workplace setting, with p < 0.01 for each. The authors should clarify whether these values refer to univariable analyses or the adjusted regression model and ensure that the Abstract is consistent with the final adjusted results.
Regression model (P9, L206–209): Please recheck the reported combination of F = 2.26, p < 0.01, and 33.4% explained variance. These values do not appear readily compatible given the number of predictors included in the model. Please report the numerator and denominator degrees of freedom, R², adjusted R², and the complete model test.
Units of the regression coefficients in Table 7: Table 6 appears to present mean domain scores of approximately 1–5, whereas Table 7 reports coefficients such as 1.27 and describes them as mean differences. If a 1–5 mean score was used, a difference of 1.27 points would represent a very large difference. If a summed score was used, the interpretation would be different. Please specify whether the outcome was a mean domain score or a summed score, provide its possible range, and clearly state the unit represented by the “Effect size” column.

Please also specify whether the reported coefficients are unstandardized regression coefficients, standardized coefficients, adjusted mean differences, or another type of estimate. The same scoring metric should be used consistently in the Methods, Tables 6 and 7, Results, and Discussion.

A full verification of the dataset, scoring procedures, tables, statistical output, Abstract, and narrative Results is recommended.

4. Ambiguity in the Definition and Measurement of AI Use

Location: Methods (P3, L97–101) and Discussion (P10, L228–230).
Issue: Participants were asked only whether they were currently using AI in clinical practice, using a Yes/No response. The manuscript does not report collecting information about the specific AI tools used, frequency of use, level of clinical integration, or tasks for which AI was used. Nevertheless, the Discussion states that participants used AI for “administration, education, or practice purposes.” These categories are not supported by the Results presented.
Requirement: If information about specific AI tools, frequency, or purposes was collected, these findings should be described in the Methods and reported in the Results. If such information was not collected, the unsupported descriptions should be removed from the Discussion.

The inability to distinguish among generative AI tools, documentation or administrative applications, educational tools, diagnostic support systems, and other forms of AI should also be acknowledged as a limitation. These technologies involve substantially different clinical risks and require different forms of training. Therefore, the finding that 53.2% of participants used AI in clinical practice should be interpreted cautiously.

5. Limited Reliability of the SHAIP Preparedness Dimension

Location: Methods (P3, L102–112).
Issue: The Cronbach’s alpha coefficient for the preparedness dimension was 0.610 for the Indonesian version and 0.632 for the original version. Although there is no universally applicable cutoff for Cronbach’s alpha and the coefficient is influenced by the number of items, a value of 0.610 indicates limited internal consistency and should not be described without qualification as demonstrating “good reliability.”
Requirement: Please revise the description of the reliability of the preparedness dimension and explain the criteria used to interpret the alpha coefficients. The limited internal consistency of this dimension should be acknowledged in the Limitations section because measurement error may have reduced the ability to identify associations with participant characteristics.

The manuscript should also provide further information regarding the translation, cultural adaptation, and validation procedures used for the Indonesian-language version of the SHAIP questionnaire. If these procedures have been reported elsewhere, the relevant validation study should be cited.

6. Sampling Method and Generalizability

Location: Methods (P3, L80–87), Results (P4, L139–146), and Table 1 (P4–5).
Issue: Participants were recruited using non-probability sampling through an online survey. In addition, 59.9% of the sample resided in Greater Jakarta or elsewhere on Java, while several other regions were represented by very small numbers of participants. The statement that participants were recruited “across Indonesia” may therefore overstate the geographical representativeness of the sample.
Requirement: Please provide, where available, the number and types of institutions invited, the number of individuals approached, the recruitment channels used, and the response rate. The potential for selection bias, digital-access bias, and regional underrepresentation should be discussed more explicitly.

The online recruitment strategy may have preferentially reached professionals with greater digital literacy, stronger interest in AI, or closer connections to the participating institutions. Statements generalizing the findings to Indonesian mental health professionals as a whole should therefore be moderated.

7. Ethical Approval Date

Location: Methods (P4, L119–121) and Institutional Review Board Statement (P12, L320–322).
Issue: The ethical approval protocol number is reported, but the date of approval is not provided. Because data collection took place between February and March 2026, readers should be able to confirm that ethical approval was obtained before recruitment and data collection began.
Requirement: Please report the date of ethical approval in both the Methods and the Institutional Review Board Statement.

Minor Comments

Replace “subjects” with “participants” throughout the manuscript.
Use consistent terminology for the study population. The manuscript alternates among “mental health professionals,” “mental healthcare workers,” “healthcare professionals,” and “providers.”
Consider adding “mental health professionals” to the keywords to improve consistency with the title and enhance discoverability.
Clearly define all abbreviations at first use and verify consistency between the main text and the Abbreviations section, including AI, ADHD, GP, LLM, NLP, SHAIP, and WHO.
Use “and” rather than an ampersand in formal prose, including references to “child and adolescent mental health services.”
Ensure consistent use of past tense when describing completed methods and analyses.
Use consistent and statistically appropriate terminology for univariable, bivariable, and multivariable analyses.
The manuscript would benefit from careful English-language editing to correct grammatical errors, awkward phrasing, punctuation errors, and inconsistencies in terminology.

Author Response

Dear Reviewer,

I really appreciate your insightful feedback for enhancing the manuscript. We have revised our manuscript accordingly and present a detailed response to your remarks.

Comment 1: Inconsistency in the Definition of the Study Population

Location: Title (P1, L2–3), Abstract (P1, L14–16), and Methods (P3, L81–86) versus Table 2 (P5, L155), Table 3 (P6, L169), and Results (P7, L179–180).
Issue: The title, abstract, and eligibility criteria broadly define the target population as Indonesian mental health professionals. However, Tables 2 and 3 and part of the Results section refer specifically to “child and adolescent mental health services.” The Methods do not indicate that participants were required to work in child and adolescent mental health services.
Requirement: Please clarify whether the study examined the general mental health workforce or specifically professionals working in child and adolescent mental health services. If the study was limited to the latter group, the title, abstract, eligibility criteria, and interpretation of the findings should be revised accordingly. If it was not limited to this group, references to child and adolescent mental health services should be removed or corrected throughout the manuscript.
In addition, four medical students were included in the sample and subsequent analyses (Table 1, P5; Table 3, P6; and Table 7, P10). Undergraduate medical students would not ordinarily be classified as mental health professionals or healthcare practitioners. Please explain the rationale and eligibility criteria for including medical students. The authors should also consider conducting a sensitivity analysis excluding these participants to confirm that their inclusion does not affect the findings.

Response 1: I appreciate the correction and apologize these inconsistencies. We corrected the phrase “child and adolescent mental health services” to “mental health services in Indonesia” throughout the text to clarify that the research studied the wider mental health workforce in Indonesia. Concerning the incorporation of medical students as mental health practitioners, we included students who were in their final year of clinical training. In Indonesia, medical students at this level frequently engage with patients and receive comprehensive training in mental health practice under the continuous supervision of psychiatrist consultants. This clarification has been incorporated into Section 3.1 (P4-5, L166-170).

Comment 2: Misinterpretation of the SHAIP Preparedness Items

Location: Abstract (P1, L22–23), Table 5 (P8, Items 6 and 8), Results (P8, L189–192), and Discussion (P11, L268–271).
Issue: The Discussion states that more than half of the participants believed that they were “personally prepared” for the introduction of AI. However, Item 6 asks whether healthcare professionals overall are prepared for the introduction of AI technology; it does not assess the respondent’s personal preparedness. Item 8 asks whether respondents believe that they have been adequately trained to use AI specific to their role. Only 21.0% agreed with Item 8, whereas 44.2% disagreed and 34.8% were neutral.
The statement in the Abstract that “most reported feeling personally unprepared” is also not directly supported by the questionnaire. The survey assessed formal training history, perceived adequacy of role-specific training, and perceptions of profession-wide preparedness, but it did not appear to directly assess personal readiness as a separate construct.

Requirement: Please distinguish clearly among:

receipt of formal AI education or training;
perceived adequacy of role-specific AI training;
perceived preparedness of healthcare professionals overall; and
personal readiness to use AI.

Response 2: I sincerely appreciate you putting this to consideration. We agree with this comment and have modified the precise language to ensure that it is more relevant to the questionnaire item. We have revised the term “personal preparedness” to “overall healthcare professional preparedness”. We have also revised the sentence “most reported feeling personally unprepared to use AI specific to their role” into “while only 21.0% agreed that they had received adequate AI training specific to their roles.” in the Abstract (P1, L25-26); the sentence “almost half (44.2%) also did not believe that they have been adequately trained to use AI that is specific to their role” into “only 21.0% agreed that they had received adequate formal training to use AI specific to their roles, while 44.2% disagreed and 34.8% were neutral.” In the Results, Section 5.1 (P8, L222-224); and also revised the sentences “In regards to preparedness for AI, more than half of participants believed they personally were prepared for the introduction of AI into their practice. Nonetheless, 82.8% had received no training or coursework on AI use in daily practice and a majority confirmed they had not been adequately trained to use AI specific to their role.” into “Regarding preparedness for AI, more than half of the participants believed that healthcare professionals were overall prepared for the introduction of AI into their practice. Nonetheless, 82.8% reported having received no formal training or coursework on AI use in daily practice and only one-fifth agreed that they had received adequate training to use AI specific to their roles” in the Discussion, Section 4.1 (P11, L311-315).

Comment 3: Numerical and Statistical Inconsistencies

Several numerical and statistical inconsistencies require clarification and correction.

Comment 3.1: Table 5, Item 10 (P8): The reported numbers total 229 participants (57 + 35 + 137), rather than the full sample of 267. The percentages total only 85.7%. Please clarify whether 38 responses were missing or whether this is a reporting error. If responses were missing, explain why this item had substantially more missing data than the other items and state whether participants were permitted to skip individual questions.
Because Item 10 contributes to the preparedness dimension, the Methods should also explain how these missing responses were handled when calculating the domain score and conducting subsequent analyses. Please state whether complete-case analysis, person-mean substitution, multiple imputation, or another method was used.

Response 3.1: Thank you for noticing this error. We have corrected Item 10 on Table 5. The total data of the “Agree continuum” column of Item 10 was previously wrong (137), we have rechecked our data and the correct total data for “Agree continuum” on Item 10 is 175 and the total participants is 267 (57 + 135 + 175 = 267). The rest of Item 10 total data is already correct (Disagree continuum and neutral, each 57 and 35 respectively). We apologize for this error.

Comment 3.2: Table 3 (P6): For participants aged 41–50 years, 10 answered “Yes” and 39 answered “No.” Given a total of 49 participants, the percentage answering “No” should be 79.6%, rather than the reported 58.4%. Please verify all numbers, percentages, and denominators presented in the tables.

Response 3.2: We have corrected the percentage to 79.6%, and apology for the mistake.

Comment 3.3: Abstract versus multivariable analysis: The Abstract states that age group, profession, and years of work experience were significantly associated with AI use (P1, L18–19). However, the multivariable model indicates that profession was the only variable with a statistically significant overall association. Age group was not significant overall (p = 0.06), although the 41–50-year category differed significantly from the reference category, and work experience was not significant after adjustment. Please clearly identify which findings are derived from univariable analyses and which are derived from the adjusted multivariable model.
Similarly, the Abstract states that perceived professional impact differed significantly by gender and workplace setting, with p < 0.01 for each. The authors should clarify whether these values refer to univariable analyses or the adjusted regression model and ensure that the Abstract is consistent with the final adjusted results.

Response 3.3: We have corrected the Abstract section and replaced the previously unmentioned univariate analysis with the multivariate analysis. We have also mentioned that the results on the Abstract are derived from multivariate analysis

Comment 3.4: Regression model (P9, L206–209): Please recheck the reported combination of F = 2.26, p < 0.01, and 33.4% explained variance. These values do not appear readily compatible given the number of predictors included in the model. Please report the numerator and denominator degrees of freedom, R², adjusted R², and the complete model test.

Response 3.4: We have corrected this error on P9 and provided other relevant data regarding the regression model: The model was statistically significant (F(14, [residual df]) = 2.26, p < 0.01), however it only explained 6.2% of the variance (adjusted R² = 0.062; R² = 0.112)

Comment 3.5: Units of the regression coefficients in Table 7: Table 6 appears to present mean domain scores of approximately 1–5, whereas Table 7 reports coefficients such as 1.27 and describes them as mean differences. If a 1–5 mean score was used, a difference of 1.27 points would represent a very large difference. If a summed score was used, the interpretation would be different. Please specify whether the outcome was a mean domain score or a summed score, provide its possible range, and clearly state the unit represented by the “Effect size” column.
Please also specify whether the reported coefficients are unstandardized regression coefficients, standardized coefficients, adjusted mean differences, or another type of estimate. The same scoring metric should be used consistently in the Methods, Tables 6 and 7, Results, and Discussion. A full verification of the dataset, scoring procedures, tables, statistical output, Abstract, and narrative Results is recommended.

Response 3.5: Reviewer 1 also pointed out this error. We have revised the analysis by calculating the mean SHAIP Impact score for each participant, rather than using the summed total score. The results of Table 7 have been analyzed using the corrected mean score, and the corresponding text on Section 3.4 has also been updated (highlighted with yellow).

Comment 4: Ambiguity in the Definition and Measurement of AI Use

Location: Methods (P3, L97–101) and Discussion (P10, L228–230).
Issue: Participants were asked only whether they were currently using AI in clinical practice, using a Yes/No response. The manuscript does not report collecting information about the specific AI tools used, frequency of use, level of clinical integration, or tasks for which AI was used. Nevertheless, the Discussion states that participants used AI for “administration, education, or practice purposes.” These categories are not supported by the Results presented.
Requirement: If information about specific AI tools, frequency, or purposes was collected, these findings should be described in the Methods and reported in the Results. If such information was not collected, the unsupported descriptions should be removed from the Discussion.

Response 4: I appreciate your input. We are aware that the original Discussion contained descriptions of AI use that were not explicitly supported by the information collected through the questionnaire. In order to prevent any potential misunderstandings, we have made revisions to the unsupported descriptions of AI use for "administration, education, or practice purposes" in the Discussion, Section 4.1 (P10, L268-269). In addition, we acknowledge this limitation in the revised manuscript. In particular, we observe in the Limitations, Section 4.3 (P12, L357-362) that the questionnaire did not differentiate between various categories of AI applications.

Comment 5: Limited Reliability of the SHAIP Preparedness Dimension

Location: Methods (P3, L102–112).
Issue: The Cronbach’s alpha coefficient for the preparedness dimension was 0.610 for the Indonesian version and 0.632 for the original version. Although there is no universally applicable cutoff for Cronbach’s alpha and the coefficient is influenced by the number of items, a value of 0.610 indicates limited internal consistency and should not be described without qualification as demonstrating “good reliability.”
Requirement: Please revise the description of the reliability of the preparedness dimension and explain the criteria used to interpret the alpha coefficients. The limited internal consistency of this dimension should be acknowledged in the Limitations section because measurement error may have reduced the ability to identify associations with participant characteristics.

Response 5: We have put the information regarding translation, cultural adaptation and validation of SHAIP questionnaire into Indonesian language on page 3-4 as follows

To ensure linguistic equivalence and cultural applicability within Indonesian context, a rigorous translation and back-translation procedure was followed, adhering to the standard guidelines for cross-cultural adaptation of self-report measures (Beaton et al., 2000). First, two bilingual psychiatrists translated the original scale into Indonesian language independently. A synthesized version was then agreed upon after resolving minor discrepancies. Second, two independent translators (blind to the original scale) back-translated the synthesized version into English. Semantic equivalence was verified by comparing the back-translation with the original source. Third, an expert panel—comprising psychiatrist, medical doctor, psychologist, nurse, reviewed all translated versions alongside the original instrument. This committee systematically evaluated semantic, idiomatic, and conceptual equivalence to resolve any cross-cultural ambiguities, thereby generating the pre-final Indonesian version. Finally, cognitive interviews were conducted with a purposive sample of 10 health workers from the target demographic to pilot test the tool. This step aimed to ensure that the target population comprehended the items exactly as intended.

In the present study, the Indonesian adaptation of SHAIP has been validated. Content validity test from six mental health professionals revealed that I-CVI (Content Validity Index for Items) and CVR (Content Validity Ratio) value of Indonesian version of SHAIP questionnaire yielded significant results (>0.79). The average content validity ratio was 0.90 and S-CVI/Ave value was 0.95, indicating good validity of the instruments. The Indonesian version of SHAIP also demonstrated acceptable reliability for the professional impact dimension (α = 0.710) with relatively lower internal consistency for the preparedness dimension (α = 0.610), comparable to the original SHAIP questionnaire (α = 0.832 for professional impact of AI and α = 0.632 for preparedness for AI).

We also revised the description of the reliability results to provide a more cautious interpretation into “The Indonesian version of SHAIP has been validated and had good reliability on both dimensions. The Indonesian version had α = 0.710 for professional impact of AI and α = 0.610 for preparedness for AI dimension, which were similar to the original version of SHAIP questionnaire (α = 0.832 for professional impact of AI and α = 0.632 for preparedness for AI).” on the Materials and Methods (P4, L131-135).

Comment 6: Sampling Method and Generalizability

Location: Methods (P3, L80–87), Results (P4, L139–146), and Table 1 (P4–5).
Issue: Participants were recruited using non-probability sampling through an online survey. In addition, 59.9% of the sample resided in Greater Jakarta or elsewhere on Java, while several other regions were represented by very small numbers of participants. The statement that participants were recruited “across Indonesia” may therefore overstate the geographical representativeness of the sample.
Requirement: Please provide, where available, the number and types of institutions invited, the number of individuals approached, the recruitment channels used, and the response rate. The potential for selection bias, digital-access bias, and regional underrepresentation should be discussed more explicitly.

Response 6: Thank you for pointing out this concern. Distribution of medical professionals in Indonesia is a common topic of discussion in the country. For example, around 48.8% of all registered general practitioners (GPs) in Indonesia are located in the Java Island (Greater Jakarta included). Additionally, the population in Java island alone make up around 56% of total Indonesia population. Considering this context, our data which 59.9% of samples resided in Greater Java is still roughly proportional to the maldistribution of medical professionals in Indonesia. We have added an explanation of this Java population proportion in Section 3.1 (P5, L173-175). Regarding the type of institutions invited, we invited private and governmental institutions across Indonesia. Recruitment channels was already mentioned (via email), whereas the number of individuals approached is not available as we approached the institutions, not individuals.

Reference data from the Indonesian Medical Association: https://www.idionline.org/article/menelusuri-permasalahan-persebaran-dokter-di-indonesia-perspektif-dokter-usia-40-tahun

Comment 7: Ethical Approval Date

Location: Methods (P4, L119–121) and Institutional Review Board Statement (P12, L320–322).
Issue: The ethical approval protocol number is reported, but the date of approval is not provided. Because data collection took place between February and March 2026, readers should be able to confirm that ethical approval was obtained before recruitment and data collection began.
Requirement: Please report the date of ethical approval in both the Methods and the Institutional Review Board Statement.

Response 7: For the ethical approval, we have added the date of approval, which is 12 January 2026 on both the Methods, Section 2.5 (P4, L142) and the Institutional Review Board Statement (P13, L390).

Comment 8: Minor Comments

Replace “subjects” with “participants” throughout the manuscript.
Use consistent terminology for the study population. The manuscript alternates among “mental health professionals,” “mental healthcare workers,” “healthcare professionals,” and “providers.”
Consider adding “mental health professionals” to the keywords to improve consistency with the title and enhance discoverability.
Clearly define all abbreviations at first use and verify consistency between the main text and the Abbreviations section, including AI, ADHD, GP, LLM, NLP, SHAIP, and WHO.
Use “and” rather than an ampersand in formal prose, including references to “child and adolescent mental health services.”
Ensure consistent use of past tense when describing completed methods and analyses.
Use consistent and statistically appropriate terminology for univariable, bivariable, and multivariable analyses.
The manuscript would benefit from careful English-language editing to correct grammatical errors, awkward phrasing, punctuation errors, and inconsistencies in terminology.

Response 8: Thank you for the revision. We have also corrected the manuscript accordingly in response to the reviewers’ comments (please refer to the new manuscript file). English-language editing has been conducted with assistance from English translator in order to correct grammatical mistakes, awkward constructions, punctuation inaccuracies, and terminological discrepancies.

Once again thank you very much for your comprehensive feedback, I really appreciate it.

Reviewer 3 Report

Comments and Suggestions for Authors

This cross-sectional survey study examines awareness, usage patterns, and perceptions of AI among 267 Indonesian mental health professionals, using the validated SHAIP questionnaire. Its aim is to characterize the gap between AI adoption and AI training/preparedness in this workforce, and to identify demographic and professional factors, such as age, profession, gender, and workplace setting, that shape both AI use and perceptions of its professional impact. The paper positions this gap as a patient-safety and ethics issue, arguing for structured AI education to be integrated into mental health professional development in Indonesia.

The title and most of the paper refer to "mental health professionals", but the Methods and Table 2 heading reference "child and adolescent mental health service", while the sample and questionnaire don't appear age-group-specific. This inconsistency should be resolved throughout (title, tables, and text) to accurately reflect the study population.
The preparedness subscale shows relatively low internal consistency (α = 0.610–0.632), which is below conventional acceptability thresholds (~0.70). The authors should explicitly discuss this as a limitation, since it affects confidence in the "no significant predictors of preparedness" finding.
Since no variables reached p<0.20 in univariate analysis for the preparedness dimension, no multivariate model was built, but this is a substantive finding (i.e., preparedness is uniformly low regardless of demographics) that deserves more discussion rather than simply being noted as a lack of analysis. Consider explaining what this null result implies practically.
Non-probability/convenience sampling via institutional email distribution likely introduces selection bias (e.g., toward more tech-engaged or urban/Java-based respondents, who made up ~60% of the sample). The Limitations section should explicitly discuss potential response bias and the skewed geographic/professional distribution (e.g., few psychologists, medical students, or participants from eastern Indonesia).
Some p-values are reported ambiguously (e.g., "p=0.06" appearing in both Table 6 for profession under two different dimensions with different values, or "Age group... p = 0.06" in Table 4 juxtaposed against individual category p-values of 0.01/0.37/0.80, which reads confusingly). Standardize decimal precision, clarify overall vs. per-category significance tests, and ensure Table/text values match exactly (e.g., double-check the reliability α values, which are listed twice with different numbers in section 2.3 without clear labeling of which set refers to the original vs. Indonesian-validated version).

Author Response

Dear Reviewer,

We appreciate your valuable input to enhance our article. Herein, we present a point-by-point answer to your comments:

Comment 1: The title and most of the paper refer to "mental health professionals", but the Methods and Table 2 heading reference "child and adolescent mental health service", while the sample and questionnaire don't appear age-group-specific. This inconsistency should be resolved throughout (title, tables, and text) to accurately reflect the study population.

Response 1: We appreciate your feedback and apologize for the discrepancies in our table titles. We have revised all instances throughout the manuscript, including the title, Methods, table headings, and in-text references, to consistently use "mental health professionals" / "mental health service," which accurately reflects our study population and questionnaire scope (not only child and adolescent mental health service).

Comment 2: The preparedness subscale shows relatively low internal consistency (α = 0.610–0.632), which is below conventional acceptability thresholds (~0.70). The authors should explicitly discuss this as a limitation, since it affects confidence in the "no significant predictors of preparedness" finding.

Response 2: We agree that this is a significant limitation to recognize. We have added the following to the Limitations section (4.3): "Additionally, the preparedness subscale in this study demonstrated relatively low internal consistency (α = 0.610), below the conventional threshold of 0.70, consistent with the comparatively lower reliability of this subscale in the original SHAIP instrument as well (α = 0.632). This suggests items within this subscale may not fully capture a unified construct, possibly reflecting the smaller number of items in this dimension. This may partly explain why no demographic or professional variables were significantly associated with preparedness scores, as the observed null association could reflect measurement imprecision rather than a genuine absence of relationship."

Comment 3: Since no variables reached p<0.20 in univariate analysis for the preparedness dimension, no multivariate model was built, but this is a substantive finding (i.e., preparedness is uniformly low regardless of demographics) that deserves more discussion rather than simply being noted as a lack of analysis. Consider explaining what this null result implies practically.

Response 3: We agree and have expanded our discussion of this finding rather than treating it as a mere absence of analysis. In section 4.1 (Discussion), we now state: "Consistent with this, perceived preparedness did not differ significantly by any demographic examined, including age, sex, profession, workplace, or years of experience. Given that formal AI training was low across the entire sample, low preparedness appears to be a shared condition rather than one concentrated within any particular subgroup. This suggests that gaps in AI readiness among psychiatric healthcare professionals are more likely driven by systemic factors, such as the general absence of institutional training infrastructure, than by individual-level characteristics like age or clinical experience." This reframes the null result as evidence of a broadly shared training gap with practical implications for institution-wide (rather than subgroup-targeted) intervention.

Comment 4: Non-probability/convenience sampling via institutional email distribution likely introduces selection bias (e.g., toward more tech-engaged or urban/Java-based respondents, who made up ~60% of the sample). The Limitations section should explicitly discuss potential response bias and the skewed geographic/professional distribution (e.g., few psychologists, medical students, or participants from eastern Indonesia).

Response 4: We agree with that and we have expanded the limitations section to explicitly address the selection bias issue. “Subsequently, sample recruitment via email distribution would introduce selection bias and limit generalizability, as participants more familiar with technology (particularly younger respondents) may have been more likely to participate in the study.”

Comment 5: Some p-values are reported ambiguously (e.g., "p=0.06" appearing in both Table 6 for profession under two different dimensions with different values, or "Age group... p = 0.06" in Table 4 juxtaposed against individual category p-values of 0.01/0.37/0.80, which reads confusingly). Standardize decimal precision, clarify overall vs. per-category significance tests, and ensure Table/text values match exactly (e.g., double-check the reliability α values, which are listed twice with different numbers in section 2.3 without clear labeling of which set refers to the original vs. Indonesian-validated version).

Response 5: We have addressed this in three ways. First, we clarified the reliability values reported in section 2.3 by explicitly labeling which α values correspond to the original SHAIP instrument versus the Indonesian-validated adaptation. Second, we added explanatory footnotes to Tables 6 and 7 clarifying that reported p-values represent overall (omnibus) tests of association versus individual regression coefficients, respectively. Third, we restructured Table 4 to separate "Overall p" (likelihood ratio test for each factor) from "Category p" (individual coefficient significance relative to the reference group), resolving the ambiguity between the age-group overall p-value (p = 0.06) and its category-level p-values (0.01, 0.37, 0.80). We also cross-checked all p-values and confidence intervals reported in the text against their corresponding tables.

Once again thank you very much for your feedback.

Reviewer 4 Report

Comments and Suggestions for Authors

Please review commentaries included in the manuscript and check for spaces before parenthesis of references.

Comments for author File: Comments.pdf

Author Response

Dear Reviewer,

Thank you for your valuable inputs to our manuscript, we really appreciate it. We provide here the point-by-point responses to your comments.

Comment 1: Typographical errors in some parts throughout the manuscript: P1 (L33-34), P2 (L46, 49, 51, 56, 62, 68), P10 (L234, 243-244, 247, 251), P10 (L257, 267, 272, 278, 283, 286)

Response 1: We have made the necessary typhographical corrections on each of the pages you provided. Additionally, we have conducted a thorough evaluation to confirm that the typing format is accurate.

Comment 2: Page 6: Need to fix this table to match number and percentage with Resident category and so on.

Response 2: Thank you for pointing this out. We have corrected the display in Table 3 (page 6) to ensure that each occupation corresponds accurately with its frequency and percentage.

Comment 3: Page 8: Need to explain what these two numbers mean

Response 3: We have put the remarks about these numbers on the note written under the Table 6 on page 8:

Note. Values are presented as M ± SD. p-values for each demographic factor represent overall test of association between the factor and each perception dimension.

Comment 4: Page 10: Considering these are Muslim countries, it should also considered the role of women in the society and their access to education.

Response 4: We agree with this perspective as it is also articulated in the referenced paper. This aspect has been incorporated into the relevant paragraph. (P11, Line 284-285)

Thank you

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

Thank You for providing a revised version. The paper improved substantially. However, there is remaining clarifications to made::

final-year medical students (n=4, 1.5\%) and "other health professionals" (n=21, 7.9\%) are included in a survey targeted at mental health practitioners. Medical students are not licensed practitioners, which generate confusion into clinical usage claims. Consider running a sensitivity analysis excluding the medical students to see if regression outcomes remain stable.
In Table 4: Age group as an overall factor shows p=0.06, yet the 41–50 age bracket shows significance (p=0.01, OR 0.28). Please clarify whether the overall model significance accounts for multi-category testing corrections.
Table 7: The linear regression model for the Professional Impact dimension explains only 6.2% of the total variance. While statistically significant (F = 2.26, p < 0.01), this low predictive power should be explicitly highlighted in the discussion as a major limitation, indicating that unmeasured systemic or individual factors account for 93.8% of the perception variance.
Please ensure terms like "y.o." in tables (e.g., Table 3, Table 6) are spelled out in footers or main body text.
The study evaluates AI broadly. Adding a brief discussion on how different types of AI (e.g., administrative LLMs vs. diagnostic risk-prediction algorithms) might alter clinician perception would strengthen the clinical relevance of Section 4.3.

Author Response

Dear Reviewer,

I really appreciate your constructive feedback. We have revised the manuscript accordingly.

Comment 1: Final-year medical students (n=4, 1.5\%) and "other health professionals" (n=21, 7.9\%) are included in a survey targeted at mental health practitioners. Medical students are not licensed practitioners, which generate confusion into clinical usage claims. Consider running a sensitivity analysis excluding the medical students to see if regression outcomes remain stable.

Response 1: Concerning final year medical students, they were engaged in the mental outpatient unit during the survey period. In Indonesia, medical students at this level exposed with patients and actively participate in mental health care, under the supervision of psychiatric consultants. Although a final-year medical student is not officially recognized as a mental health professional, they often address and handle mental health issues in clinical setting under supervision. We additionally include their responses to provide insights from academics and individuals practicing within the mental health unit.

Regarding "other health professionals," they were allied practitioners from the Indonesian mental health sector, including therapists and counselors who constantly engaged in mental health services. We have included clarification on P4 L163, 166-170, and on Table 1 concerning the "Others" profession.

Comment 2: In Table 4: Age group as an overall factor shows p=0.06, yet the 41–50 age bracket shows significance (p=0.01, OR 0.28). Please clarify whether the overall model significance accounts for multi-category testing corrections.

Response 2: The significant p-value for the 41–50-year age group (OR = 0.28, p = 0.01) represents the category-specific comparison within the reference group (≤40 years). The other age categories did not exhibit statistical significance. No multiple-comparison corrections were made to the category-specific analysis, as these were regression coefficients comparing each prespecified age category with the reference category.

Comment 3: Table 7: The linear regression model for the Professional Impact dimension explains only 6.2% of the total variance. While statistically significant (F = 2.26, p < 0.01), this low predictive power should be explicitly highlighted in the discussion as a major limitation, indicating that unmeasured systemic or individual factors account for 93.8% of the perception variance.

Response 3: Thank you for pointing this out. We have added this concern as limitation in P13 L 371-375.

Comment 4: Please ensure terms like "y.o." in tables (e.g., Table 3, Table 6) are spelled out in footers or main body text.

Response 4: Thank you for your revision. This has been incorporated into the footer of Table 3 and Table 6 (y.o. = years old).

Comment 5: The study evaluates AI broadly. Adding a brief discussion on how different types of AI (e.g., administrative LLMs vs. diagnostic risk-prediction algorithms) might alter clinician perception would strengthen the clinical relevance of Section 4.3.

Response 5: We have added a short discussion on P13 L 358-360.

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you for carefully addressing my previous comments. Most of the issues have been satisfactorily resolved. I have only two remaining points from my previous review:

In Section 3.4, the denominator degrees of freedom remain as a placeholder: “F(14, [residual df]) = 2.26.” Please replace “[residual df]” with the actual value.
For SHAIP Item 10, Table 5 correctly reports 175 participants (65.5%) in the agree continuum, but the narrative Results still state 51.3%. Please correct this to 65.5%.

Apart from these minor corrections, I have no further comments.

Author Response

Dear Reviewer,

I really appreciate your valuable feedback. We have revised the manuscript according to your suggestion.

Comment 1: In Section 3.4, the denominator degrees of freedom remain as a placeholder: “F(14, [residual df]) = 2.26.” Please replace “[residual df]” with the actual value

Response 1: Thank you for your feedback. We have added the actual value of residual df (P10, L241)

Comment 2: For SHAIP Item 10, Table 5 correctly reports 175 participants (65.5%) in the agree continuum, but the narrative Results still state 51.3%. Please correct this to 65.5%.

Response 2: Thank you for noticing this mistake, we have corrected the percentage (P9, L226).