Medication Adherence and Risks of Mortality and End-Organ Damage in Asian Patients with Type 2 Diabetes: A Cohort Study from Southern Taiwan

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

 

Reviewer comments: 

The manuscript addresses an important clinical question but suffers from major methodological and analytical limitations that substantially weaken the validity of its conclusions. Significant concerns regarding adherence measurement, confounding control, statistical power, and internal consistency of results preclude acceptance in its current form.

1. Medication adherence was measured only once at baseline using a self-reported tool; how do the authors justify drawing five-year prognostic conclusions without accounting for adherence changes over time?

2. Given the strong age imbalance across adherence groups (nearly 9 years difference), why was age not more rigorously handled (e.g., stratification or interaction analysis) to rule out residual confounding?

3. The study claims adequate power for mortality but admits limited power for other endpoints; how can null findings for cardiovascular and renal outcomes be interpreted as meaningful rather than false negatives?

4. Why were causes of death not analyzed, especially when the main finding is all-cause mortality and the discussion speculates extensively on non-cardiovascular causes?

5. Baseline LDL and HbA1c levels already differed across adherence groups; how do the authors exclude reverse causality, where worse metabolic control leads to poorer adherence rather than the opposite?

6. Antidiabetic and lipid-lowering therapies were not standardized or adjusted for in the models; how can medication class, intensity, and changes during follow-up be ruled out as major confounders?

7. The hazard ratio reported for hyperlipidemia and myocardial infarction (HR ~1.08) appears implausibly precise for this sample size; can the authors clarify variable coding and model specification?

8. Several confidence intervals are extremely wide, crossing clinically meaningful ranges; why do the authors still draw strong interpretive conclusions from these unstable estimates?

9. The MMAS-8 is treated as a categorical exposure without sensitivity analyses using continuous scores; how robust are the findings to alternative adherence definitions?

10. Given the single-center design, exclusion of patients lost to follow-up, and reliance on chart-reviewed outcomes, how do the authors address the high risk of selection bias and outcome misclassification?

Author Response

The manuscript addresses an important clinical question but suffers from major methodological and analytical limitations that substantially weaken the validity of its conclusions. Significant concerns regarding adherence measurement, confounding control, statistical power, and internal consistency of results preclude acceptance in its current form.

Comment 1: Medication adherence was measured only once at baseline using a self-reported tool; how do the authors justify drawing five-year prognostic conclusions without accounting for adherence changes over time?

Response 1: Thank you for this important comment. Medication adherence was assessed once at baseline and treated as a fixed exposure variable for risk stratification. Our objective was to evaluate the prognostic significance of baseline adherence status rather than the effect of sustained adherence over time.

We agree that changes in adherence during follow-up were not captured and may influence long-term outcomes. We have therefore revised the Discussion Section to clarify this point and explicitly acknowledge it as a study limitation.

 

Comment 2: Given the strong age imbalance across adherence groups (nearly 9 years difference), why was age not more rigorously handled (e.g., stratification or interaction analysis) to rule out residual confounding?

Response 2: Thank you for raising this important concern. Upon careful review of the original dataset, we identified a typographical error in the reported mean age of the good adherence group. The correct mean age is 65.95 ± 9.61 years, not 69.95 ± 9.61 years as previously stated. We sincerely apologize for this error and have corrected the value in the revised manuscript.

Even after correction, patients in the good adherence group remained statistically older than those in the moderate and poor adherence groups (p < 0.05). Given the observational design of our study, baseline differences are expected. To address potential confounding, age was included as a covariate in statistical analysis. The association between poor adherence and all-cause mortality remained significant after adjustment for age and other prespecified covariates. Therefore, the primary findings are unlikely to be explained by age-related confounding.

 

Comment 3: The study claims adequate power for mortality but admits limited power for other endpoints; how can null findings for cardiovascular and renal outcomes be interpreted as meaningful rather than false negatives?

Response 3: Thank you for the important question. We agree that the study may have been underpowered for individual cardiovascular and renal endpoints due to the relatively small number of events. Therefore, the absence of statistically significant associations for nonfatal cardiovascular and renal outcomes should be interpreted cautiously and does not exclude the possibility of type II error.

We have revised the Discussion Section (lines 264–266) to explicitly state “the follow-up duration and event rates may have limited the ability to detect differences in outcomes characterized by slow development, such as ESKD and nonfatal cardiovascular events.” in order to clarify that these findings are exploratory rather than definitive evidence of no association.

 

Comment 4: Why were causes of death not analyzed, especially when the main finding is all-cause mortality and the discussion speculates extensively on non-cardiovascular causes?

Response 4: We thank the reviewer for this important suggestion. We have now summarized the causes of death in Appendix Table A1. Malignancy and infection accounted for the majority of deaths, whereas cardiovascular causes were less frequent. The Discussion Section has been revised to reflect these findings and to avoid speculative interpretation, with lines 227-230 stating the following: “Among the 26 deaths observed in our cohort, the most common cause was malignancy (n=11, 42.3%), followed by infections (n=8, 30.8%). Detailed causes of death are presented in Appendix Table A1.”

 

Comment 5: Baseline LDL and HbA1c levels already differed across adherence groups; how do the authors exclude reverse causality, where worse metabolic control leads to poorer adherence rather than the opposite?

Response 5: Thank you for this important comment. We acknowledge that reverse causality is a potential concern in observational studies. In our design, medication adherence was assessed at baseline, and metabolic outcomes were evaluated prospectively over five years, establishing temporal precedence.

Baseline LDL and HbA1c levels were included as covariates in the multivariable analyses. Thus, the associations observed reflect differences in follow-up metabolic control beyond initial baseline values.

We recognize that bidirectional relationships between metabolic control and adherence behaviors are possible, and we have clarified in the section on limitations that our findings indicate association rather than definitive causation, with lines 268-269 stating, “Moreover, reverse causality between metabolic control and adherence behaviors could not be fully excluded in this observational study.”

 

Comment 6: Antidiabetic and lipid-lowering therapies were not standardized or adjusted for in the models; how can medication class, intensity, and changes during follow-up be ruled out as major confounders?

Response 6: Thank you for raising this important question. Given the observational design, residual confounding by differences in pharmacotherapy cannot be fully excluded. Our findings therefore reflect associations observed under real-world treatment conditions rather than the isolated effects of specific medication regimens. We have clarified this limitation in the revised manuscript and moderated the interpretation accordingly on lines 270-272: “Third, differences in medication class, intensity, and adjustments during follow-up were not explicitly modeled, and residual confounding related to pharmacotherapy may remain.”

 

Comment 7: The hazard ratio reported for hyperlipidemia and myocardial infarction (HR ~1.08) appears implausibly precise for this sample size; can the authors clarify variable coding and model specification?

Response 7: Thank you for this important methodological comment. Hyperlipidemia was coded as a binary variable (0 = absence, 1 = presence) based on documented diagnosis at baseline. Cox proportional hazards models were performed using SPSS with myocardial infarction as the dependent variable and prespecified covariates entered simultaneously.

The reported hazard ratio (HR ~ 1.08) reflects the relative risk associated with the presence of hyperlipidemia compared with its absence after multivariable adjustment. We have rechecked the variable, covariates, and reference categories and confirm that the estimates were correctly generated. The narrow confidence interval reflects the relatively high prevalence of hyperlipidemia in the cohort and the number of events observed.

 

Comment 8: Several confidence intervals are extremely wide, crossing clinically meaningful ranges; why do the authors still draw strong interpretive conclusions from these unstable estimates?

Response 8: Thank you for this important comment. We acknowledge that several confidence intervals were wide, reflecting the limited number of events for certain cardiovascular and renal outcomes, which indicate statistical imprecision and reduced power to detect modest effect sizes.

We did not intend to overinterpret these findings. Rather, our primary conclusion emphasizes the robust association between poor adherence and all-cause mortality. We respectfully note that, in the manuscript, these imprecise estimates are not interpreted as definitive evidence but are rather presented descriptively with appropriate caution.

 

Comment 9: The MMAS-8 is treated as a categorical exposure without sensitivity analyses using continuous scores; how robust are the findings to alternative adherence definitions?

Response 9: Thank you for this thoughtful comment. We categorized MMAS-8 scores according to established and validated cutoff values commonly used in prior studies, allowing clinically meaningful interpretation of adherence levels. While alternative modeling approaches, such as treating MMAS-8 scores as continuous variables, may provide additional nuance, our primary objective was to compare clinically recognized adherence categories.

 

Comment 10: Given the single-center design, exclusion of patients lost to follow-up, and reliance on chart-reviewed outcomes, how do the authors address the high risk of selection bias and outcome misclassification?

Response 10: We acknowledge that the single-center design may limit generalizability and introduce potential selection bias. However, participants were consecutively enrolled from routine outpatient care, which reduces the likelihood of selective inclusion based on adherence status or outcome risk. Major endpoints such as all-cause mortality and ESKD are relatively objective outcomes and less susceptible to misclassification.

Nonetheless, we recognize that reliance on chart review may introduce some degree of outcome misclassification, and this has been noted as a limitation. In the revised manuscript, we have added the following on lines 272-275: “Fourth, as a single-center study, generalizability may be limited. Fifth, exclusion of patients lost to follow-up may introduce selection bias, and residual attrition bias cannot be entirely excluded. Sixth, outcomes were ascertained through medical record review; therefore, some degree of misclassification may be possible.”

Reviewer 2 Report

Comments and Suggestions for Authors

In this paper, Chen et al. studied the association between baseline patient-reported medication adherence and long-term mortality, cardiovascular, renal, and metabolic outcomes in adults with type 2 diabetes followed for five years in Southern Taiwan.

While the topic is clinically relevant, the degree of originality is limited. Similar associations between adherence, particularly measured by the Morisky Medication Adherence Scale, and mortality or cardiovascular outcomes have been extensively reported in prior literature, including studies conducted in Asian populations. The present study does not clearly articulate a novel hypothesis, nor does it provide a sufficiently differentiated conceptual framework to distinguish its contribution from existing evidence. The regional focus on Southern Taiwan is not adequately leveraged to generate new insights, as no region-specific healthcare, behavioral, or system-level factors are rigorously analyzed.

Several methodological concerns weaken the validity of the conclusions. Medication adherence was assessed only at baseline, despite a prolonged follow-up period of five years. Given that adherence is a time-varying behavior, this approach introduces a high risk of exposure misclassification and limits causal interpretation. Additionally, antidiabetic and lipid-lowering treatments were not standardized, monitored longitudinally, or incorporated into the analytical models, making it difficult to disentangle adherence effects from treatment intensity or medication changes over time.

The exclusion of participants with incomplete data or loss to follow-up is insufficiently described, and the absence of a detailed flow or attrition analysis raises concerns regarding potential selection bias. Moreover, key confounding variables, such as socioeconomic status, educational level, mental health conditions, and lifestyle factors, were not assessed or adjusted for, despite their known associations with both medication adherence and mortality.

The statistical approach is constrained by limited event numbers, particularly for nonfatal cardiovascular and renal outcomes. Although the authors acknowledge that the study is powered only to detect large effect sizes (hazard ratios ≥ 3.0), multiple endpoints and subgroup analyses are nonetheless performed. This substantially increases the risk of false-negative findings and undermines confidence in the reported null associations.

No adjustment for multiple comparisons is applied, despite the analysis of several primary and secondary outcomes. Furthermore, baseline differences between adherence groups, most notably in age and HbA1c, are considerable, yet the adequacy of multivariable adjustment is questionable given the low events-per-variable ratio in the Cox models.

The results section contains conceptual and factual weaknesses. Baseline differences in LDL cholesterol across adherence groups are not adequately accounted for when interpreting longitudinal lipid outcomes, raising concerns about residual confounding. Similarly, the lack of observed differences in HbA1c and renal function may reflect insufficient sensitivity of the study design rather than true equivalence. There is also a critical error in wording: all-cause mortality is described as being “lower significantly in the poor compliance group,” which contradicts the presented data (language error: incorrect comparative direction; should be “significantly higher”). Such inaccuracies reduce confidence in the reliability of data reporting.

The discussion tends to overinterpret the findings, particularly by framing medication adherence as a global marker of health engagement without directly measuring or adjusting for relevant behavioral or healthcare utilization variables. While plausible, this interpretation remains speculative.

Limitations are acknowledged but not sufficiently integrated into the interpretation of results. For example, the low number of renal and cardiovascular events and the single-time-point adherence assessment are mentioned, yet the discussion continues to imply a true lack of association between adherence and these outcomes. Alternative explanations, such as inadequate statistical power, competing risks, or misclassification bias, are not explored in sufficient depth.

Comments on the Quality of English Language

The manuscript contains multiple English language and editorial issues that require substantial revision. These include grammatical errors, inconsistent terminology, and typographical mistakes (for example: “the strength and consistency of these associations remain incompletely understood and has not been fully elucidated”, “Haazard ratio” in Table 4, “Moresby Medication Adherence Scale” instead of “Morisky”, “albumin albumin-creatinine ratio”, and many others). Overall, the manuscript would benefit from thorough professional language editing to improve clarity, precision, and consistency.

Author Response

In this paper, Chen et al. studied the association between baseline patient-reported medication adherence and long-term mortality, cardiovascular, renal, and metabolic outcomes in adults with type 2 diabetes followed for five years in Southern Taiwan.

Comment 1: While the topic is clinically relevant, the degree of originality is limited. Similar associations between adherence, particularly measured by the Morisky Medication Adherence Scale, and mortality or cardiovascular outcomes have been extensively reported in prior literature, including studies conducted in Asian populations. The present study does not clearly articulate a novel hypothesis, nor does it provide a sufficiently differentiated conceptual framework to distinguish its contribution from existing evidence. The regional focus on Southern Taiwan is not adequately leveraged to generate new insights, as no region-specific healthcare, behavioral, or system-level factors are rigorously analyzed.

Several methodological concerns weaken the validity of the conclusions. Medication adherence was assessed only at baseline, despite a prolonged follow-up period of five years. Given that adherence is a time-varying behavior, this approach introduces a high risk of exposure misclassification and limits causal interpretation.

Response 1: Medication adherence was assessed once at baseline and treated as a fixed exposure variable for risk stratification. Our objective was to evaluate the prognostic significance of baseline adherence status rather than the effect of sustained adherence over time.

We agree that changes in adherence during follow-up were not captured and may influence long-term outcomes. We have therefore revised lines 265-268 in the Discussion Section, adding “Second, medication adherence was assessed only at baseline and may have changed over time. Because adherence is a dynamic behavior, a single assessment may not accurately represent long-term exposure.” to clarify this point and explicitly acknowledge it as a study limitation.

 

Comment 2: Additionally, antidiabetic and lipid-lowering treatments were not standardized, monitored longitudinally, or incorporated into the analytical models, making it difficult to disentangle adherence effects from treatment intensity or medication changes over time.

Response 2: Given the observational design, residual confounding by differences in pharmacotherapy cannot be fully excluded. Our findings therefore reflect associations observed under real-world treatment conditions rather than the isolated effects of specific medication regimens. We have clarified this limitation in the revised manuscript and moderated the interpretation accordingly on lines 270-272: “Third, differences in medication class, intensity, and adjustments during follow-up were not explicitly modeled, and residual confounding related to pharmacotherapy may remain.”.

 

Comment 3: The exclusion of participants with incomplete data or loss to follow-up is insufficiently described, and the absence of a detailed flow or attrition analysis raises concerns regarding potential selection bias. Moreover, key confounding variables, such as socioeconomic status, educational level, mental health conditions, and lifestyle factors, were not assessed or adjusted for, despite their known associations with both medication adherence and mortality.

Response 3: Thank you for this important mention. We acknowledge that important psychosocial and socioeconomic factors, including educational level, income, mental health conditions, and lifestyle behaviors, were not systematically assessed in this study. These variables are known to influence both medication adherence and mortality risk and may represent potential sources of residual confounding. We have now explicitly acknowledged this limitation in the revised manuscript on lines 276-277: “residual confounding by unmeasured factors, including lifestyle, socioeconomic, and psychological variables, cannot be totally excluded.”

 

Comment 4: The statistical approach is constrained by limited event numbers, particularly for nonfatal cardiovascular and renal outcomes. Although the authors acknowledge that the study is powered only to detect large effect sizes (hazard ratios ≥ 3.0), multiple endpoints and subgroup analyses are nonetheless performed. This substantially increases the risk of false-negative findings and undermines confidence in the reported null associations.

Response 4: Thank you for this important comment. We acknowledge that the limited number of nonfatal cardiovascular and renal events within the five-year follow-up reduces statistical power for secondary and subgroup analyses.

The additional cardiovascular and renal outcomes were analyzed to provide broader clinical context. We have clarified in the Discussion Section that these findings should be interpreted cautiously and considered exploratory, and lines 263-265 now state the following: “the follow-up duration and event rates may have limited the ability to detect differences in outcomes characterized by slow development, such as ESKD and nonfatal cardiovascular events.”

 

Comment 5: No adjustment for multiple comparisons is applied, despite the analysis of several primary and secondary outcomes. Furthermore, baseline differences between adherence groups, most notably in age and HbA1c, are considerable, yet the adequacy of multivariable adjustment is questionable given the low events-per-variable ratio in the Cox models.

Response 5: Thank you for this important comment. All-cause mortality was prespecified as the primary endpoint and represents the main focus of inference; the cardiovascular and renal outcomes were considered exploratory, and the findings should be interpreted cautiously.

Regarding multivariable adjustment, covariates were prespecified based on clinical relevance, and the number of variables included in each Cox model was limited to reduce overfitting. We acknowledge that limited event numbers may affect precision and have emphasized this as a limitation on lines 263-265 in the Discussion Section of the revised manuscript.

 

Comment 6: The results section contains conceptual and factual weaknesses. Baseline differences in LDL cholesterol across adherence groups are not adequately accounted for when interpreting longitudinal lipid outcomes, raising concerns about residual confounding. Similarly, the lack of observed differences in HbA1c and renal function may reflect insufficient sensitivity of the study design rather than true equivalence. There is also a critical error in wording: all-cause mortality is described as being “lower significantly in the poor compliance group,” which contradicts the presented data (language error: incorrect comparative direction; should be “significantly higher”). Such inaccuracies reduce confidence in the reliability of data reporting.

Response 6: Thank you for this careful review. We sincerely apologize for the wording error in the Results Section. The statement has been corrected to “In contrast, all-cause mortality was significantly higher in the poor compliance group” on lines 171-172, consistent with the data presented. We regret this oversight and have carefully rechecked the manuscript to ensure accuracy throughout.

  Regarding baseline LDL differences, we agree that residual confounding is an important consideration. Baseline LDL levels were included as covariates in the multivariable analyses evaluating follow-up lipid outcomes to account for initial differences. We have clarified this point in the Methods and Discussion Sections to improve transparency, with “adjusting for age, sex, HbA1c, LDL, hypertension, hyperlipidemia, CKD, prior MI, and heart failure.” added on lines 122-123 in the revised manuscript.

  With respect to the changes in HbA1c and renal function, we acknowledge that the absence of statistically significant differences may reflect limited sensitivity and statistical power rather than definitive equivalence. We have revised lines 208-210 in the Discussion Section, stating “Although no statistically significant differences were observed in HbA1c or renal outcomes, the possibility of modest effects cannot be excluded due to limited statistical power” to emphasize that these findings should be interpreted cautiously.

 

Comment 7: The discussion tends to overinterpret the findings, particularly by framing medication adherence as a global marker of health engagement without directly measuring or adjusting for relevant behavioral or healthcare utilization variables. While plausible, this interpretation remains speculative.

Response 7: Thank you for this comment. We have revised lines 276-277 in the Discussion Section, stating “Lastly, residual confounding due to unmeasured factors, including lifestyle, socioeconomic, and psychological variables, could not be completely excluded.” to moderate this interpretation and to clarify that such mechanisms remain speculative.

 

Comment 8: Limitations are acknowledged but not sufficiently integrated into the interpretation of results. For example, the low number of renal and cardiovascular events and the single-time-point adherence assessment are mentioned, yet the discussion continues to imply a true lack of association between adherence and these outcomes. Alternative explanations, such as inadequate statistical power, competing risks, or misclassification bias, are not explored in sufficient depth.

Response 8: On lines 263-265 in the revised Discussion Section, we have altered the language to avoid implying a definitive lack of association and have emphasized that inadequate statistical power, potential competing risks, and possible outcome misclassification may have contributed to the null findings: “the follow-up duration and event rates may have limited the ability to detect differences in outcomes characterized by slow development, such as ESKD and nonfatal cardiovascular events.” These considerations have now been incorporated into the interpretation of the results.

 

Comment 9: The manuscript contains multiple English language and editorial issues that require substantial revision. These include grammatical errors, inconsistent terminology, and typographical mistakes (for example: “the strength and consistency of these associations remain incompletely understood and has not been fully elucidated”, “Haazard ratio” in Table 4, “Moresby Medication Adherence Scale” instead of “Morisky”, “albumin albumin-creatinine ratio”, and many others). Overall, the manuscript would benefit from thorough professional language editing to improve clarity, precision, and consistency.

Response 9: The manuscript has been thoroughly revised to correct these issues and improve clarity and consistency. In addition, the revised version will undergo professional English language editing to ensure accuracy and readability prior to resubmission.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Review report

I would like to thank the authors for their efforts in responding to the comments of the reviewers and for incorporating their valuable suggestions into the manuscript. I recommend that the manuscript be accepted in its present form. However, I would like to request the authors to carefully review the entire manuscript once again before its publication to ensure that all aspects are checked.

Author Response

Reviewer 1

I would like to thank the authors for their efforts in responding to the comments of the reviewers and for incorporating their valuable suggestions into the manuscript. I recommend that the manuscript be accepted in its present form. However, I would like to request the authors to carefully review the entire manuscript once again before its publication to ensure that all aspects are checked.

 

Reply:

We sincerely thank the reviewer for the positive evaluation of our revised manuscript and for acknowledging our efforts in addressing the previous comments. We greatly appreciate your constructive feedback throughout the review process.

In accordance with your suggestion, we have carefully reviewed the entire manuscript once again to ensure clarity, consistency, accuracy, and completeness prior to publication.

Thank you again for your thoughtful review and support of our work.

Reviewer 2 Report

Comments and Suggestions for Authors

The authors did not address the major concern raised in the first round of review.

Comments on the Quality of English Language

The manuscript contains multiple English language and editorial issues that require substantial revision. These include grammatical errors, inconsistent terminology, and typographical mistakes (for example: “the strength and consistency of these associations remain incompletely understood and has not been fully elucidated”, “Haazard ratio” in Table 4, “Moresby Medication Adherence Scale” instead of “Morisky”, “albumin albumin-creatinine ratio”, and many others). Overall, the manuscript would benefit from thorough professional language editing to improve clarity, precision, and consistency.

Author Response

Reviewer 2

The manuscript contains multiple English language and editorial issues that require substantial revision. These include grammatical errors, inconsistent terminology, and typographical mistakes (for example: “the strength and consistency of these associations remain incompletely understood and has not been fully elucidated”, “Haazard ratio” in Table 4, “Moresby Medication Adherence Scale” instead of “Morisky”, “albumin albumin-creatinine ratio”, and many others). Overall, the manuscript would benefit from thorough professional language editing to improve clarity, precision, and consistency.

 

Reply:

Comment 1: The manuscript contains multiple English language and editorial issues that require substantial revision.

 

Response1: We appreciate the reviewer’s concern regarding language quality. Following the first round of review, the manuscript underwent professional English language editing through MDPI Author Services.

In addition, after receiving the present comment, we have once again carefully reviewed the entire manuscript to ensure grammatical accuracy, consistency of terminology, and elimination of typographical errors.

 

Comment 2:“The strength and consistency of these associations remain incompletely understood and has not been fully elucidated.”

 

Response 2: We thank the reviewer for citing this sentence. We would like to clarify that the subject–verb agreement issue (“remain … has”) had already been corrected in the revised manuscript submitted after the first review round. The sentence currently reads:“the strength and consistency of these associations have not been fully elucidated and remain unclear.”

We have rechecked the current version to confirm that the corrected wording is retained.

 

Comment 3:“Haazard ratio” in Table 4

 

Response 3: We appreciate the reviewer for noting this typographical issue. This typographical error was corrected to “Hazard ratio” in the revised manuscript submitted after the first review round. We have verified that the corrected spelling appears in Table 4 of the current version.

 

Comment 4:“Moresby Medication Adherence Scale” instead of “Morisky”

 

Response 4: Thank you for drawing attention to this point. This terminology error was also corrected in the revised manuscript after the first round of review. The correct term, “Morisky Medication Adherence Scale,” is consistently used in the current version. We have rechecked the manuscript to confirm consistency.

 

Comment 5:“albumin albumin-creatinine ratio”

 

Response 5: We appreciate the reviewer’s careful reading. This duplication error had likewise been corrected in the previously revised manuscript. The term now consistently appears as “urinary albumin-to-creatinine ratio.” We have verified that no duplication remains.

 

We sincerely appreciate the reviewer’s emphasis on language quality. While the specific examples cited had already been corrected in the revised submission and the manuscript had undergone professional language editing, we have nonetheless performed another comprehensive review to ensure clarity, precision, and consistency prior to publication.

 

We are grateful for the reviewer’s careful evaluation and for the opportunity to further refine our work.

Round 3

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have not adequately addressed the primary concern that was raised during the first round of peer review. Specifically, the key issue identified by the reviewers remains unresolved, and the revised manuscript does not provide sufficient clarification, additional analysis, or substantive modifications to respond to this concern. As a result, the main point of criticism continues to limit the validity, clarity, or overall contribution of the work. A more thorough and direct response to this issue is necessary in order to satisfactorily address the reviewers’ comments.

Comments on the Quality of English Language

I propose to have the manuscript checked by a native English speaking person.

Author Response

The authors have not adequately addressed the primary concern that was raised during the first round of peer review. Specifically, the key issue identified by the reviewers remains unresolved, and the revised manuscript does not provide sufficient clarification, additional analysis, or substantive modifications to respond to this concern. As a result, the main point of criticism continues to limit the validity, clarity, or overall contribution of the work. A more thorough and direct response to this issue is necessary in order to satisfactorily address the reviewers’ comments.

Reply: We thank the reviewer for the continued evaluation of our manuscript. When revising the manuscript, we carefully reconsidered the key issues identified by the reviewers and implemented several substantive revisions aimed at improving the clarity, methodological transparency, and interpretation of the study.

 

In particular, we revised the manuscript to more clearly frame medication adherence as a baseline prognostic factor, acknowledging that adherence was measured only once at study entry. The interpretation of the findings has been adjusted accordingly throughout the Abstract, Results, and Discussion sections at line 36-39 “Poor medication adherence as assessed at baseline may be related to a higher risk of all-cause mortality and poorer lipid control, while no statistically significant differences were observed for nonfatal cardiovascular or renal outcomes”, line 193-196 “In this prospective cohort study, baseline medication adherence was not associated with a statistically significant difference in the risk of nonfatal myocardial infarction, stroke, heart failure, and progression to ESKD. However, patients with poor adherence at baseline exhibited a higher risk of all-cause mortality during follow-up”, and line 294-297 “In conclusion, while poor medication adherence at baseline in patients with T2DM was not significantly associated with increased risks of nonfatal myocardial infarction, stroke, heart failure, or ESKD, it may be related to a higher risk of all-cause mortality and poorer lipid control” respectively. We also expanded the Limitations section to emphasize the implications of single-time-point adherence measurement and the possibility that adherence behaviors may change over time at line 264-269 “Second, medication adherence was assessed only once at baseline and may have changed during the follow-up period. Because adherence behavior is dynamic, a single baseline measurement may not fully capture long-term adherence patterns. Therefore, the present findings should be interpreted as reflecting the prognostic significance of baseline adherence status, rather than the sustained effects of adherence over time”.

 

Regarding baseline treatment categories, the primary aim of this study was to evaluate the prognostic significance of patient-reported medication adherence in a real-world clinical setting, rather than to assess the comparative effects of specific antidiabetic drug classes. In routine clinical practice, treatment regimens in our cohort were frequently individualized and modified during follow-up according to glycemic control, comorbidities, and physician judgment. Because medication regimens often involved combinations of agents and changed over time, treatment exposure could not be modeled consistently across the entire follow-up period.

 

Our analysis therefore focused on adherence behavior as a patient-level factor reflecting engagement with long-term diabetes management, independent of specific pharmacologic regimens. Nevertheless, we agree that differences in medication classes or treatment intensity may introduce residual confounding. We have now clarified this point in the Methods on lines 106-111 “Antidiabetic treatment during follow-up was determined by treating physicians according to routine clinical practice. No protocol-mandated restrictions were placed on the choice of glucose-lowering medications, and treatment regimens could be adjusted at the clinicians’ discretion based on individual patient needs. However, detailed information regarding specific medication classes, treatment intensity, and dose adjustments during follow-up was not systematically recorded in the study dataset” and emphasized it as a limitation of the study on lines 271-276 “Third, detailed information on specific medication classes, treatment intensity, and dose adjustments during follow-up was not systematically available in our dataset. Therefore, we were unable to adjust for differences in pharmacotherapy patterns across adherence groups. Because medication regimens may influence outcomes such as mortality, lipid control, glycemic changes, and renal function, residual confounding related to pharmacologic treatment cannot be excluded”.

 

To further improve transparency, we have added p-values for baseline comparisons across adherence groups in Table 1. We have also revised the Results and Discussion section to briefly summarize the baseline comparisons on lines 163-166 “Baseline HbA1c levels differed significantly across adherence groups and were highest in the poor adherence group (6.90±0.97% in the good adherence group, 6.99±0.96% in the moderate adherence group, and 7.32±1.21% in the poor adherence group; p = 0.007)”, 238-249 “Baseline HbA1c levels were already higher in the poor adherence group compared with the other adherence categories. However, the changes in HbA1c over the five-year follow-up period did not differ significantly across groups. Because our analysis did not include strict adjustment for multiple behavioral and clinical factors that may influence glycemic control, these findings should be interpreted cautiously. Nevertheless, the observed pattern may suggest that individuals with poorer medication adherence tend to have less optimal glycemic control. At the same time, glycemic outcomes in patients with T2DM are influenced by multiple factors beyond pharmacotherapy alone, including dietary patterns, physical activity, insulin titration practices, and structured care programs such as shared care initiatives for diabetes, which are commonly implemented in Taiwan. These systemic and behavioral influences may attenuate the measurable impact of medication adherence on glycemic outcomes”, and 251-254 “Notably, LDL cholesterol levels tended to be higher in patients with poorer adherence both at baseline and at the end of the five-year follow-up, suggesting a consistent pattern of less optimal lipid control among individuals with lower adherence”. We also removed the analyses previously presented in Table 4, as they were potentially difficult to interpret given the limited number of events. This modification has helped simplify the presentation of the results and improve the clarity of the statistical reporting.

 

Finally, we revised portions of the Discussion on lines 233-237 in the revised manuscript to ensure that certain interpretations—such as the suggestion that adherence may reflect broader health engagement—are explicitly framed as hypothesis-generating rather than direct inferences from the data. The revised manuscript now reads “However, this interpretation should be considered hypothesis-generating, as important factors such as lifestyle behaviors, healthcare access, psychosocial factors, education, and mental health were not directly measured in the present study. Future studies incorporating these variables may help clarify the mechanisms underlying the observed associations.”

 

We hope these revisions adequately address the concerns raised during the first review round and improve the clarity and rigor of the manuscript. We sincerely appreciate the reviewer’s continued feedback, which has helped us strengthen the presentation and interpretation of our work.