## 1. Overall assessment

This manuscript addresses a genuinely under-served question. Sleep disturbance among stroke survivors in Southeast Asia is poorly characterised, RISKESDAS 2018 is an appropriate data source, and an analytic sample of 7,147 stroke survivors is a real asset. The authors have responded constructively to several points from the first review round: the terminology in the title has been changed, a confidence interval has been added to the overall prevalence estimate, the survey-design limitation is now acknowledged, Hosmer-Lemeshow statistics are reported, and several of the impossible odd's ratios in v1 have been fixed.

## 2. Major issues

### 2.1 Table 1: the male education rows are transposed

As printed, Table 1 states that 781 of 932 male participants with sleep disturbance (83.8%) had university-level education, compared with 329 of 2,642 (12.5%) without sleep disturbance. This cannot be correct:

| Check | Value |
|---|---|
| University level, total column | 605 |
| University level, sum of four cells as printed | 781 + 329 + 54 + 150 = 1,314 |
| University level, sum after swapping the two male rows | 72 + 329 + 54 + 150 = 605 |
| Non-university level, total column | 5,810 |
| Non-university level, sum after the same swap | 781 + 2,145 + 995 + 1,889 = 5,810 |

The swap is confirmed independently by Table 3. Using the corrected cells, the crude odds ratio for university level in males is (72/329)/(781/2,145) = 0.601, which reproduces Table 3 exactly. For no formal education, (79/168)/(781/2,145) = 1.291, which also reproduces Table 3 exactly. With the printed cells, neither value is recoverable.

The published table therefore reports the exact opposite of the finding described in the text, which states that higher educational attainment is protective. This should be corrected and the entire table rebuilt from the source output rather than edited cell by cell.

### 2.2 Table 3: the confidence interval for university-level education in males is wrong

From the corrected counts (72/329 versus 781/2,145), the standard error of the log odds ratio is 0.137, giving a 95% CI of approximately 0.460 to 0.786. Table 3 reports 0.456 to 0.848. The printed interval is not symmetric on the log scale and cannot arise from a Wald calculation.

For comparison, the corresponding female row (0.683, 0.496 to 0.942) reproduces exactly from the counts, as does the male no-formal-education row (1.291, 0.976 to 1.708). The error is isolated to this one cell, which is precisely the kind of residual error that a full re-check against the output would have caught.

### 2.3 Table 5: the age confidence interval in males is uninterpretable, and contradicts the author response

Table 5 reports an adjusted OR for age in males of 1.006 with a 95% CI of 0.995 to 1.010. The point estimate is not contained symmetrically within the interval on the log scale, and the interval is not centred on the estimate at all. In the response to Reviewer 1, comment 6, the authors state that this interval "should read 1.000 to 1.013." The published table does not show that. Please reconcile the correction letter with the table.

### 2.4 The reported interaction p-values are inconsistent with the stratum-specific estimates

Section 3.3 states that formal interaction tests were significant for heart disease (p < 0.01) and asthma (p = 0.02). These values cannot be reconciled with the adjusted estimates in Table 6.

| Comorbidity | Male adjusted OR (95% CI) | Female adjusted OR (95% CI) | Implied interaction OR | Implied z | Implied p |
|---|---|---|---|---|---|
| Heart disease | 1.282 (0.997 to 1.648) | 1.615 (1.287 to 2.025) | 1.26 | 1.34 | ~0.18 |
| Asthma | 1.297 (0.937 to 1.795) | 1.658 (1.226 to 2.243) | 1.28 | 1.09 | ~0.28 |

Repeating the calculation with the crude estimates from Table 4 gives p values of approximately 0.14 and 0.34. Neither reaches the reported significance. Either the interaction models differ substantially from the stratified models in ways not described, or the p-values are in error.

This matters because the interaction analysis is the only formal evidence offered for the paper's central claim of sex differences. Please report the interaction analysis in full: model specification, covariates, sample size, interaction odds ratios with confidence intervals, and p-values, ideally as a supplementary table.

Relatedly, interaction terms are reported only for the two comorbidities that favor females. The male-specific findings emphasised in the Abstract (daily smoking, education, rural residency, physical activity) are supported only by the difference in statistical significance between the two stratified models, which is not a test of difference. Interaction terms should be reported for every variable on which a sex-difference claim is made, or the claims should be reworded as sex-specific descriptive findings.

### 2.5 The headline sex difference is never formally tested

The prevalence of 34.4% in females versus 26.1% in males is the paper's principal result. It appears in the title framing, the Abstract, Section 3.2, the Discussion, and the Conclusions, yet no chi-square statistic, p-value, prevalence ratio, prevalence difference, or confidence interval for the difference appears anywhere in the manuscript. Reviewer 2 raised exactly this point in round 1 and it has not been addressed. Please report sex-specific prevalence with confidence intervals and a formal comparison and consider reporting an age-standardised comparison given the mean age difference of roughly three years between sexes.

### 2.6 The reference numbering does not match the reference list

The reference list is ordered alphabetically by first author, but the in-text citation numbers appear to follow the original submission order. The result is that the numbered citations point to the wrong sources throughout. Examples:

| In-text claim | Cited number | Reference list entry at that number | Apparently intended source |
|---|---|---|---|
| Meta-analysis of 22 studies, post-stroke prevalence 38.2% | 1 | Baglioni 2014, polysomnographic meta-analysis in insomnia disorder | Hasan 2021, Stroke (list no. 8) |
| Higher risk of cognitive impairment after stroke; 8.2% incidence in Taiwan | 3 | Bull, Global Physical Activity Questionnaire | Hasan 2023, Behav Sleep Med (list no. 9) |
| Female predominance in the Chinese general population | 6 | Ferre, strokes and sleep disorders | Cao 2017, PLoS One (list no. 4) |
| GPAQ reliability in Indonesia, kappa 0.44 to 0.78 | 7 | Fulk, sleep problems and quality of life | Bull, GPAQ (list no. 3) |
| Female predominance among American adults | 9 | Hasan 2023, cognitive impairment | Zuo 2022, Sleep Health (list no. 16) |

Every citation should be checked and renumbered. Two knock-on issues:

- The GPAQ psychometric claim is currently attributed to a paper on stroke and quality of life. Even if the intended source is Bull et al., that is a nine-country reliability and validity study, not an Indonesian validation. If no Indonesia-specific validation exists, the sentence should say so.
- Five of the seventeen references are by the corresponding author. That is defensible in a narrow field, but the introduction and discussion would be strengthened by broader engagement with the sleep literature.

### 2.7 The terminology revision is incomplete

The authors state that "post-stroke insomnia" and "insomnia" have been replaced throughout. They have not been. Remaining instances:

- Section 3.2: "Figure 2 depicts the comparative prevalence of PSI among males and females."
- Discussion: "The stroke's site, including the right hemisphere, thalamus, or brainstem, is associated with PSI"; "may exacerbate post-stroke sleep disturbance."
- Conclusions: "the prevalence of PSI remains high over time"; "Females have a higher prevalence of PSI compared to males."
- Keywords: "post-stroke insomnia" and "insomnia" both retained.
- Figure 1: exclusion box reads 'missing data in "insomnia" variable.'
- Figure 2: y-axis reads "Insomnia Prevalence."

Since the Conclusions are what most readers will carry away, retaining PSI there undermines the entire revision.

### 2.8 The Table 2 denominator change from 5,366 to 7,147 is not explained

In response to Reviewer 1, the authors state that Table 2 has been "corrected to reflect the full sample of 7,147 participants based on the SPSS output." An increase of 1,781 participants is not a transcription correction. Similarly, the female prevalence moved from 31.2% to 34.4%, which corresponds to roughly 115 participants changing outcome status, and cannot be a rounding or abstraction slip.

Please state plainly whether the analytic dataset changed between versions, and if so, what changed and why.

A related problem: every row of Tables 1 and 2 now sums exactly to 7,147, implying zero missingness on all fourteen covariates. This contradicts Section 2.6, which states that complete-case analysis was used and that participants with missing covariate values were excluded from the respective regression models. Either there is no missingness, in which case complete-case analysis is moot and the statement should be removed, or there is missingness, in which case it is not being shown. Please report variable-level missingness and the analytic n for each regression model, as promised in Section 2.6 but not delivered anywhere in the Results.

### 2.9 The described modelling strategy contradicts the tables

Section 2.6 specifies forward stepwise selection with entry at p < 0.05 and removal at p < 0.10. Tables 5 and 6 nonetheless report every candidate variable, including several that could never have entered under those criteria: age in males (p = 0.55), marital status in males (p = 0.89), employment in males (p = 0.31), and alcohol in females (p = 0.46).

The footnotes compound the problem. Table 5 says the model "includes all variables presented in the table simultaneously," Table 6 says it "includes all significant variables from univariate analysis (p < 0.05) entered simultaneously," and both then say a forward stepwise method was used. Simultaneous entry and stepwise selection are mutually exclusive. Furthermore, Table 6 contains cancer, CKD, and alcohol, none of which reached p < 0.05 in Table 4, contradicting its own footnote.

Please state what was actually fitted. If a full model was used, say so and remove the stepwise language, which would be the preferable choice here given that stepwise selection with a large n and pre-specified covariates offers little benefit and distorts inference.