Depression and Suicidal Ideation Among Men who have sex with men and Transgender Women in Chiang Mai, Thailand

Jul 4^th, 2026

Dear reviewer,

Thank you for all of your comments. Please find our responses below.

The authors address a vital and clinically relevant public health intersection: mental health disparities among sexual and gender-diverse populations and engagement with HIV prevention services. The use of culturally validated psychometric instruments, such as the Thai versions of the PHQ-9 and the C-SSRS, is a notable methodological strength. However, there are significant conceptual, statistical, and presentational flaws that must be resolved to ensure the scientific validity and rigor of the findings.

Below is a detailed breakdown of the required revisions.
I. Conceptual Framework and Literature Review

• Behaviour vs. Identity Conflation: In the introduction, the manuscript groups "men who have sex with men (MSM), transgender women (TGW), and other Lesbian, Gay, Bisexual, Transgender, Queer or Questioning, and other sexual and gender minority identities". This demonstrates an epidemiological conflation. "MSM" is a behavioural categorization used to describe transmission risk networks, whereas LGBTQ+ terms denote psychosocial sexual orientations and gender identities. The text must be revised to clearly distinguish between behavioural risk groups and sexual/gender minority identities.

Response: Thank you for pointing out this important issue. We have revised the text in the 1^st paragraph of the introduction to make it clear, as follows.

“Men who have sex with men (MSM) and transgender women (TGW) are HIV key populations, classified respectively by sexual behavior and gender identity rather than by sexual orientation. Alongside these behaviorally and demographically defined groups, sexual and gender minority populations more broadly, encompassing lesbian, gay, bisexual, transgender, queer, and questioning (LGBTQ+) identities, experience a substantially higher burden of mental health problems than the general population.”

• Pooling MSM and TGW Data: The manuscript evaluates MSM and TGW as a single entity. While public health frameworks frequently group MSM and TGW together as "key populations" due to shared epidemiological HIV vulnerabilities, their social determinants of mental health are vastly different. A transgender woman's experience of minority stress differs fundamentally from the homophobic stigma experienced by cisgender MSM. The authors must either stratify their regression models and descriptive statistics by gender identity (MSM vs. TGW) or explicitly justify in the limitations section why pooling them was necessary (e.g., sample size constraints), acknowledging the specific vulnerabilities this aggregation obscures.

Response: We absolutely agree with the reviewer. Although MSM and TGW were analyzed together as key populations, we acknowledge that these groups experience different forms of minority stress and stigma. Stratified analyses were not performed because the sample size of TGW participants (24(16%)) was insufficient to support reliable subgroup regression models. Future studies with larger TGW samples should examine these populations separately. This issue is addressed in the study limitation at the paragraph before the last of the revised manuscript as follows.
“Third, MSM and TGW were analysed together because the number of TGW participants was insufficient to support reliable subgroup analyses. These populations experience distinct forms of minority stress, and future studies with larger samples should examine them separately.”

• Outdated Literature on India: The authors cite Tomori et al., 2016, to highlight the "elevated risk among these groups" in India. However, the socio-legal landscape in India fundamentally shifted in 2018 with the decriminalization of homosexuality. Citing pre-2018 data fails to reflect the contemporary structural context. The authors must integrate post-2018 literature showing that depression remains high due to syndemic factors despite legislative reforms.

Response: We thank the reviewer for this important observation. We have revised the Introduction by removing the pre-2018 country-specific examples and replacing them with recent systematic review evidence (2022). The revised text now cites contemporary literature demonstrating that depression and suicidal ideation remain disproportionately common among MSM and transgender women despite legal and policy advances, largely because of persistent stigma, discrimination, and minority stress. The revised text is in the 1^st paragraph read as follows.
“Recent systematic reviews further indicate that elevated rates of depression and suicidal ideation persist across diverse settings despite legal and policy advances, largely because stigma, discrimination, minority stress, and other interacting psychosocial factors continue to adversely affect mental health ( Operario et al., 2022; Pellicane & Ciesla, 2022).”

• Unsubstantiated Regional Generalizations: The manuscript claims findings "reflect broader patterns observed across Southeast Asia". Southeast Asia is highly heterogeneous regarding LGBTQ+ rights and stigma. The authors must cite specific, recent regional data to substantiate this claim.

Response: We thank the reviewer for this important and insightful comment. We agree that Southeast Asia is highly heterogeneous in terms of sociocultural norms, legal frameworks, and levels of stigma toward sexual and gender minority populations. In light of this, we have revised the manuscript to remove the generalized statement referring to “broader patterns observed across Southeast Asia,” as it was not sufficiently supported by specific regional evidence.
The revised paragraph now focuses more precisely on evidence from Thailand, particularly studies conducted in Bangkok and other local settings, to ensure contextual accuracy and relevance to the study population. This modification improves the conceptual clarity and avoids overgeneralization beyond the available empirical evidence. It is read as follows.
“In Thailand, emerging evidence indicates that LGBTQ+ populations continue to experience substantial mental-health challenges. Recent studies have highlighted high levels of psychological distress among sexual and gender minority individuals, with a survey of Thai medical students reporting notably elevated rates of depressive and anxiety symptoms among those identifying as LGBTQ+ (Vadhanavikkit et al., 2025). Research among TGW in Bangkok has also documented a high prevalence of depressive symptoms, with low perceived social support identified as a significant contributing factor (Tantirattanakulchai & Hounnaklang, 2021). Despite these concerns, systematic data on mental-health outcomes among key populations in northern Thailand remain limited, underscoring the need for focused research in this context.”

II. Statistical and Methodological Flaws

• Mathematical Circularity in Regression Models: In Table 3, the linear regression model for "Depression Severity" utilizes the "PHQ-9 Diagnosis (Cut-off≥10)" as a predictor. Because the outcome variable (continuous PHQ-9 score) and this predictor (categorized PHQ-9 score) are derived from the exact same scale, this introduces a severe error of mathematical circularity. The model must be completely re-run without this predictor.

Response: We agree that including the dichotomized PHQ-9 diagnosis as a predictor of the continuous PHQ-9 score introduced mathematical circularity because both variables were derived from the same instrument. Accordingly, we removed PHQ-9 diagnosis from the regression model and re-ran the multivariable analysis using only independent demographic, psychosocial, and PrEP-related variables. Table 3 and the corresponding Results have been revised.
Predictors
B
95% CI
p-value
Age group (40–59 years vs. 20–39 years)
-0.34
-2.27, 1.59
0.730

Age group (≥60 years vs. 20–39 years)
1.16
-6.22, 8.54
0.756

Education level
(Undergraduate or higher vs. High/Vocational school or lower)
0.36
-1.17, 1.90
0.640

Employment status (Employed vs. Unemployed/student)
0.10
-1.37, 1.57
0.896

Monthly income (Higher income vs. Low income)
-0.26
-1.56, 1.05
0.699

Experience of discrimination
(No discrimination vs. Experienced discrimination)
-1.32
-2.59, -0.06
0.040

Family/Society pressure (Low/No pressure vs. High pressure)
-0.71
-2.12, 0.71
0.325

Community safety (Not always safe vs. Always safe)
0.68
-0.57, 1.93
0.284

Current stress level (High stress vs. Low stress)
2.37
-0.49, 5.23
0.103

Social support (No support vs. Has support)
-0.97
-4.79, 2.85
0.617

LGBTQ+ community involvement (Yes vs. No)
-0.40
-1.93, 1.12
0.603

PrEP use (PrEP user vs. Non-user)
-1.77
-3.07, -0.47
0.008

 

• Severe Overfitting in Suicidal Ideation Analysis: The study reports a very low frequency of suicidal ideation: 7% among PrEP users and 4% among non-users. In a sample of 150 participants, this equates to roughly 8 total events. Running a multivariable logistic regression model with nine covariates on only 8 events guarantees severe statistical overfitting. A standard statistical rule of thumb requires at least 10 outcome events per predictor variable. This analysis is statistically invalid and should be downgraded to an exploratory bivariate analysis or omitted entirely.

Response: We thank the reviewer for this important observation. We agree that the number of suicidal ideation events (n = 8) was insufficient to support a multivariable logistic regression model without substantial risk of overfitting. In response, we have removed the logistic regression analysis and replaced it with an exploratory bivariate analysis using Fisher's exact test, which is more appropriate for sparse binary outcome data. The Results and Methods sections have been revised accordingly.
The revised text in the last paragraph of the result section is now read as follows.
“Suicidal ideation was reported by 8 of the 150 participants (5.3%). Among PrEP users, 5 participants (6.7%) reported suicidal ideation compared with 3 participants (4.0%) among non-users. Fisher's exact test showed no statistically significant association between PrEP use and suicidal ideation (p = 0.719).”
Table 4 Bivariable analysis for suicidal ideation
PrEP Use
No suicidal ideation
Suicidal ideation
Total
Non-user
72 (96.0%)
3 (4.0%)
75

PrEP user
70 (93.3%)
5 (6.7%)
75

Severe Selection and Survivor Bias: The inclusion criteria define a PrEP user as someone utilizing the medication for ≥6 months, comparing them to never-users. By explicitly excluding early dropouts (individuals who initiated PrEP but stopped within 1-5 months), the researchers introduced a massive survivor bias. Furthermore, the study attempts to correlate long-term PrEP engagement (≥6 months) with the PHQ-9, which strictly measures depressive symptoms over a brief two-week window. This misalignment must be addressed.
Response: We appreciate the reviewer's comment. Our intention was to compare two clearly defined groups—participants who had used PrEP continuously for at least six months and participants who had never used PrEP. Individuals who had previously initiated but discontinued PrEP were therefore outside the predefined comparison groups. We acknowledge that this design may introduce survivor bias because long-term PrEP users may differ systematically from those who discontinue treatment. We have clarified the study design in the Methods and added this limitation to the Discussion.
In the method section:
“To align with the programmatic eligibility criteria for PrEP services, the study population comprised individuals assigned male at birth and transgender women who reported sexual activity with men. The study was designed to compare two predefined exposure groups: participants who had used PrEP continuously for at least six months and participants who had never used PrEP. Individuals who had previously initiated but discontinued PrEP were excluded because they did not meet either predefined exposure category.“
In the limitation section:
“Second, the study compared long-term PrEP users (≥6 months) with individuals who had never used PrEP. Participants who discontinued PrEP before six months were excluded, which may have introduced survivor bias.”
"Phantom" Variables and Incomplete Models: In Table 3, "PrEP Adherence" is listed as a predictor. However, this variable is never defined or mentioned in the Methods section, and it is impossible to apply an "adherence" score to the non-user control group. The codebook shows PrEP_Adherence was indeed collected on a 4-point scale (1 = Daily, 2 = Almost daily, 3 = Sometimes, 4 = 2/6 months) and offers no value for "Not Applicable" or "Non-user." Therefore, in a regression model containing both PrEP users and non-users, how did the software handle the 75 non-users? If non-users were left blank, the statistical software would have used "listwise deletion," dropping all 75 non-users from the Table 3 regression model entirely. Additionally, the Data Analysis section explicitly claims covariates like monthly income, community safety, and LGBTQ+ involvement were entered into all multivariable models, yet these are omitted from Table 3. The authors must report the full model accurately.
Response: We thank the reviewer for identifying this inconsistency. PrEP adherence was measured only among current PrEP users and was therefore not applicable to non-users. Including this variable in the combined regression model was inappropriate and could have resulted in listwise deletion of participants with missing adherence data. We have removed PrEP adherence from the multivariable model and revised the Methods, Results, and Table 3 accordingly. The final model now includes only variables measured consistently across all participants.
The text and Table 3 in the Result section are noe read as follows.
“PrEP use was independently associated with lower depression scores (B = −1.77, 95% CI: −3.07 to −0.47; p = 0.008). Age group, education level, employment status, monthly income, family/societal pressure, community safety, current stress level, social support, and LGBTQ+ community involvement were not significantly associated with depression scores (all p > 0.05).”
Table 3. Linear Regression for Depression Severity.
Predictors
B
95% CI
p-value
Age group (40–59 years vs. 20–39 years)
-0.34
-2.27, 1.59
0.730

Age group (≥60 years vs. 20–39 years)
1.16
-6.22, 8.54
0.756

Education level
(Undergraduate or higher vs. High/Vocational school or lower)
0.36
-1.17, 1.90
0.640

Employment status (Employed vs. Unemployed/student)
0.10
-1.37, 1.57
0.896

Monthly income (Higher income vs. Low income)
-0.26
-1.56, 1.05
0.699

Experience of discrimination
(No discrimination vs. Experienced discrimination)
-1.32
-2.59, -0.06
0.040

Family/Society pressure (Low/No pressure vs. High pressure)
-0.71
-2.12, 0.71
0.325

Community safety (Not always safe vs. Always safe)
0.68
-0.57, 1.93
0.284

Current stress level (High stress vs. Low stress)
2.37
-0.49, 5.23
0.103

Social support (No support vs. Has support)
-0.97
-4.79, 2.85
0.617

LGBTQ+ community involvement (Yes vs. No)
-0.40
-1.93, 1.12
0.603

PrEP use (PrEP user vs. Non-user)
-1.77
-3.07, -0.47
0.008

 

• Contradictory Outcome Definitions: In the Methods, the authors state that the C-SSRS responses were summarized into a "single multi-categorical variable". However, in the Data Analysis section, they claim to use a "binary logistic regression model" for suicidal ideation. This inconsistency makes the suicidality analysis unreproducible and must be clarified.

Response: We thank the reviewer for identifying this inconsistency. We agree that the original description of the C-SSRS outcome did not correspond to the reported statistical analysis. The Methods have been revised to clarify that responses to the six C-SSRS items were summarized into a binary variable indicating the presence or absence of suicidal ideation/risk, defined as endorsement of one or more C-SSRS items. Because the number of events was small, the original logistic regression analysis was removed and replaced with Fisher's exact test. The Methods and Results sections have been revised accordingly.

In the data analysis part of the method section:
“To identify factors associated with depression, multivariable linear regression analysis was performed using the continuous PHQ-9 total score as the dependent variable. Independent variables entered into the model were age group, education level, employment status, monthly income, experience of discrimination, family/societal pressure, community safety, current stress level, social support, LGBTQ+ community involvement, and PrEP use. Unstandardized regression coefficients (B), 95% confidence intervals (CIs), and p-values were reported. Model assumptions were assessed by examining residual plots for linearity and homoscedasticity, while multicollinearity was evaluated using variance inflation factors (VIFs), with values <2 indicating no evidence of problematic multicollinearity.
Because suicidal ideation/risk occurred infrequently, resulting in too few outcome events for reliable multivariable modelling, logistic regression was not performed. Instead, Fisher's exact test was used to compare suicidal ideation/risk between PrEP users and non-users.”

In the result section:
“Suicidal ideation was reported by 8 of the 150 participants (5.3%). Among PrEP users, 5 participants (6.7%) reported suicidal ideation compared with 3 participants (4.0%) among non-users. Fisher's exact test showed no statistically significant association between PrEP use and suicidal ideation (p = 0.719).”
Table 4 Bivariable analysis for suicidal ideation
PrEP Use
No suicidal ideation
Suicidal ideation
Total
Non-user
72 (96.0%)
3 (4.0%)
75

PrEP user
70 (93.3%)
5 (6.7%)
75

Inter-site Demographic Discrepancies: The study recruited from two vastly different clinical environments: PIMAN Clinic (a research-based university clinic) and MPLUS Polyclinic (a community-based Key Population-Led Health Service). University cohorts typically differ significantly from community-led cohorts in sociodemographic and baseline symptom profiles. The authors must report a comparative analysis of the two recruitment sites to rule out site-level clustering and confounding.

• Unadjusted Site-Level Clustering: Standard epidemiological practice dictates that when sampling from distinct clusters, the "site" must be controlled for in the regression models to account for facility-level variance. The failure to control for "Recruitment Site" in Table 3 leaves a massive, unmeasured confounder in the data.

Response: We thank the reviewer for the important comments regarding the two study sites. Participants were recruited equally from the two study clinics (75 participants per clinic). We added a sentence in the 1^st paragraph of the method section to specific their characteristics as follows.
“Participants in the two clinics represented the key population residing in urban area and had comparable demographics; some of them alternated between the two clinics.”
We regret to inform you that the comparative analysis of the two recruitment sites cannot be done as we did not capture this information on the questionnaire. Consequently, recruitment site could not be included as a covariate in the multivariable analyses. We have clarified this in the Methods as mentioned above, and acknowledged it as a limitation in the revised manuscript.
“Fourth, recruitment site could not be adjusted for because recruitment site information was not captured. Residual confounding related to differences between service settings therefore cannot be excluded.“

• Unmeasured Confounding (HIV Status): The inclusion criteria define non-users simply as individuals who have "never used PrEP". The manuscript fails to specify the HIV status of this group. If individuals living with HIV are included in the non-user cohort, this represents a massive, unmeasured confounding variable, as HIV status independently impacts depression. The authors must clarify the HIV status of the sample and adjust their limitations accordingly.

Response: We thank the reviewer for raising this important point. We agree that the recruitment process was insufficiently described in the original manuscript. Participants were recruited after completing their routine clinical services at HIV prevention clinics. At both study sites, HIV testing or HIV status verification is routinely performed according to standard clinical protocols before recruitment into HIV prevention services. Individuals with HIV were therefore not targeted for recruitment, and study participants were recruited from clients eligible for HIV prevention services. We have clarified this process in the Methods section. Accordingly, HIV status was not considered a source of confounding in the present study.
The clarification is in the 3^rd paragraph of the method section. It is read as follows.
“Participants were recruited after completing their routine clinic services. At both study sites, HIV testing or HIV status verification was routinely performed according to standard clinical protocols before recruitment into HIV prevention services. Consequently, recruitment was restricted to clients eligible for HIV prevention services, and individuals living with HIV were not approached for enrolment.”
III. Data Presentation and Terminology

• Misleading Descriptive Statistics in Table 1: The summary statistics in Table 1 utilize row percentages rather than standard column percentages, completely confusing the interpretation of baseline characteristics. For example, for "High stress", the table displays 0 (0.00%) for PrEP users and 7 (100%) for non-users. This indicates that of the 7 total stressed people, 100% belonged to the non-user group, rather than showing the actual prevalence of high stress within each respective cohort. Table 1 must be entirely recalculated to display column percentages.

Response:  We revised Table 1 as follows.

Table 1. Comparison of sociodemographic and psychosocial characteristics by PrEP use status
Characteristic
PrEP Users (n = 75)
Non-Users (n = 75)
p-value
Age group
0.031

20–40 years
65 (86.7%)
73 (97.3%)

After 40 years
10 (13.3%)
2 (2.7%)

Education level
0.541

High/Vocational school or lower
13 (17.3%)
17 (22.7%)

Undergraduate or higher
62 (82.7%)
58 (77.3%)

Living arrangement
0.252

Living alone
39 (52.0%)
31 (41.3%)

Living with others
36 (48.0%)
44 (58.7%)

Employment status
0.025

Unemployed/student
13 (17.3%)
26 (34.7%)

Employed
62 (82.7%)
49 (65.3%)

Monthly income
1.000

Low income (<20,000 THB or none)
48 (64.0%)
49 (65.3%)

Higher income (>20,000 THB)
27 (36.0%)
26 (34.7%)

Experience of discrimination
0.506

Experienced discrimination
47 (62.7%)
42 (56.0%)

No discrimination
28 (37.3%)
33 (44.0%)

Family/Society pressure
0.142

High pressure
16 (21.3%)
25 (33.3%)

Low/No pressure
59 (78.7%)
50 (66.7%)

Community safety
0.048

Always safe
48 (64.0%)
35 (46.7%)

Not always safe
27 (36.0%)
40 (53.3%)

Current stress level
0.013

High stress
0 (0.0%)
7 (9.3%)

Low stress
75 (100.0%)
68 (90.7%)

Social support
0.120

Has support
75 (100.0%)
71 (94.7%)

No support
0 (0.0%)
4 (5.3%)

LGBTQ+ community involvement
0.009

Yes
66 (88.0%)
52 (69.3%)

No
9 (12.0%)
23 (30.7%)


Causal Language and Reverse Causality: The PrEP user and non-user groups exhibit significant baseline differences in age, employment, perceived community safety, and stress levels. While the authors acknowledge cross-sectional limitations, the abstract and discussion lean heavily toward causality. The Discussion concludes that "engagement with HIV-prevention services might positively affect the mental health well-being of the key populations". Given the cross-sectional design and the survivor bias mentioned above, this directional leap is unwarranted. Language must be softened to explicitly describe "associations." The lower depression scores might simply reflect reverse causality: individuals with higher baseline psychosocial stability are more capable of accessing and adhering to preventative healthcare.  
Response: We thank the reviewer for this thoughtful comment. We agree that the cross-sectional design does not permit causal inference and that reverse causality is a plausible explanation for the observed association. In response, we have revised the Abstract and Discussion to consistently describe PrEP use as being associated with, rather than causing, lower depression scores. We have also expanded the Discussion to acknowledge that individuals with better baseline psychosocial functioning may be more likely to access and remain engaged in PrEP services, and that longitudinal studies are required to establish temporality.

The conclusion of the abstract is now read as follows.
“Among MSM and TGW attending HIV prevention services in Chiang Mai, PrEP use was independently associated with lower depression scores, whereas suicidal ideation/risk did not differ significantly between PrEP users and non-users. Experience of discrimination was associated with higher depression scores. Although causal inferences cannot be made because of the cross-sectional design, the findings highlight the importance of considering mental health alongside HIV prevention services for key populations.”

The 1^st paragraph of the discussion section was revised as follows.

“This study compared depression severity and suicidal ideation/risk between PrEP users and non-users among MSM and TGW attending HIV prevention services in Chiang Mai, Thailand. PrEP users had significantly lower depression scores than non-users, whereas suicidal ideation/risk was infrequent and did not differ significantly between the two groups. After adjustment for demographic and psychosocial factors, PrEP use remained independently associated with lower depression scores, while participants who had experienced discrimination reported significantly higher depression scores.”
 

• "Double Negative" Drafting Error: In the Results section (Factors associated with depression), the authors state that "PrEP use was independently negatively associated with lower depression scores". Mathematically, a negative association with a lower score implies a higher score. This phrasing is confusing and contradictory. It should be corrected for precision to read "associated with lower depression scores" or "negatively associated with depression severity."

Response: We change that sentence into the correct one regarding new result. The text before Table 3 is now read as follows.
“PrEP use was independently associated with lower depression scores (B = -1.77, 95% CI: -3.07 to -0.47, p = 0.008).”

• Outdated Terminology: The manuscript describes some participants as "biological males who identified themselves as female". This phrasing is outdated and pathologizing. It must be revised to standard, affirmative terminology, such as "individuals assigned male at birth (AMAB)" or "transgender women."

Response: We thank the reviewer for this important comment and the description was changed to “individuals assigned male at birth (AMAB) who identified as female" as suggested in the method section (study population and inclusion criteria), and the results (participant characteristics).
IV. Data Transparency and Selective Reporting

• The Omitted Stigma Scale: The study's codebook clearly outlines a comprehensive 28-item Stigma questionnaire that culminates in a continuous Stigma Score ranging from 0 to 112. However, this entire psychometric scale is completely absent from the manuscript. Omitting a fully collected 28-item scale, especially one directly relevant to the mental health and systemic stress of key populations, is a severe case of selective reporting. The authors must disclose this data or explicitly justify its exclusion (e.g., did the stigma score highly correlate with depression, thereby weakening the primary association with PrEP use?).

Response: We thank the reviewer for this important observation. A comprehensive stigma scale was collected as part of the broader questionnaire. However, the primary objective of the present study was to compare depression and suicidal ideation/risk between PrEP users and non-users and to identify demographic and psychosocial factors associated with depression. Consequently, the stigma scale was not included in the prespecified multivariable model. Exploratory analysis showed a weak-to-moderate correlation between stigma and depression (Pearson's r = 0.246, p = 0.002), suggesting that omission of this variable is unlikely to have materially altered the primary association between PrEP use and depression. We acknowledge that stigma is an important determinant of mental health and have noted in the Discussion that future studies should further examine its relationship with depression among MSM and TGW.