Estimating associations between antidepressant use and incident mild cognitive impairment in older adults with depression.

INTRODUCTION: Previous studies have provided equivocal evidence of antidepressant use on subsequent cognitive impairment; this could be due to inconsistent modeling approaches. Our goals are methodological and clinical. We evaluate the impact of statistical modeling approaches on the associations between antidepressant use and risk of Mild Cognitive Impairment (MCI) in older adults with depression.
METHODS: 716 participants were enrolled. Our primary analysis employed a time-dependent Cox proportional hazards model. We also implemented two fixed-covariate proportional hazards models-one based on having ever used antidepressants during follow-up, and the other restricted to baseline use only.
RESULTS: Treating antidepressant use as a time-varying covariate, we found no significant association with incident MCI (HR = 0.92, 95% CI: 0.70, 1.20). In contrast, when antidepressant use was treated as a fixed covariate, we observed a significant association between having ever used antidepressants and lower risk of MCI (HR = 0.40, 95% CI: 0.28, 0.56). However, in the baseline-use only model, the association was non-significant (HR = 0.84, 95% CI: 0.60, 1.17). DISCUSSION: Our results were dependent upon statistical models and suggest that antidepressant use should be modeled as a time-varying covariate. Using a robust time-dependent analysis, antidepressant use was not significantly associated with incident MCI among cognitively normal persons with depression.

Introduction

Depression and dementia are both common in older adults[1-5]. Previous studies have suggested that depression may be a risk factor for dementia[6] or may be prodromal to dementia[7-8]. Findings about associations between antidepressant use and subsequent cognitive impairment have been inconsistent. Several studies have suggested that antidepressant use is associated with increased risk of developing cognitive impairment and dementia[9-14]. Some studies have found an inverse relationship—that antidepressant use may decrease the risk of developing dementia[15-16]. In some cases researchers have differentiated among classes of antidepressants, such as tricyclic antidepressants (TCAs), monoamine oxidase inhibitors (MAOIs), selective serotonin reuptake inhibitors (SSRIs), and serotonin and noradrenaline reuptake inhibitors (SNRIs)[17-24]. Conclusive evidence on whether antidepressant use influences cognitive function is still lacking.

Methodological approaches to estimating the association between antidepressant use and cognitive decline vary in these prior studies, as do definitions of “exposure” (i.e., antidepressant use) and “outcome” (i.e., cognition). We hypothesize that the multitude of statistical procedures, the differing criteria used to define depression and the wide variety of approaches to characterizing antidepressant use by previous studies may contribute to the inconsistent estimates of the association that have obtained. Furthermore, we suggest that implementation of time-fixed models which are sensitive to immortal time bias could be a major limitation of previous work. Despite the well-documented shortcomings of these models, their use persists [25].

We have two goals of this study one clinical goal: to evaluate the association between antidepressant use and cognitive impairment and one methodological goal: to illuminate the benefits and pitfalls of different methodological approaches and apply relatively robust methods and definitions for inference. We provide a direct comparison of the results of a time-fixed model and a time-dependent model, the latter of which we suggest is the appropriate method for handling this data, acknowledging that our results would inevitably be in conflict with some others.

Methods

2.1 Data source

Data used in this study comes from the National Alzheimer’s Coordinating Center (NACC), which maintains a database representing the clinical enrollment of the 39 past and present Alzheimer’s Disease Centers (ADCs) supported by the U.S. National Institute on Aging/National Institutes of Health. In those centers, participants underwent annual evaluations according to a standardized protocol, the Uniform Data Set (UDS), described in detail elsewhere[26-27]. Many of the subjects represented in the NACC database were volunteers referred by themselves, family, or friends due to concerns about their memory. Written informed consents were obtained from all participants at individual ADCs where they were enrolled. Research using the NACC database was approved by the University of Washington Institutional Review Board.

2.2 Study sample

Participants were eligible for this study if they were aged 60 years or older, cognitively normal, and depressed at their first UDS visit. Participants also must have made at least three UDS visits from 2005 to 2016. Assessment of normal cognition and MCI was made by either a single clinician or a formal consensus panel at each ADC. Subjects were considered depressed if they met at least two of the following five criteria at entry into the study: 1) self-reported active depression in the last two years, 2) depression or dysphoria symptoms as reported by a co-participant on the Neuropsychiatric Inventory Questionnaire (NPI-Q)[28], 3) Geriatric Depression Scale-15 (GDS-15) score of at least six[29,30], 4) clinically depressed mood based on clinician interview, or 5) a clinical diagnosis of active depression based on current UDS examination and the clinician’s best judgment.

Fig 1 shows that there were 11,096 participants aged 60 years or older and cognitively normal at baseline. Among those, 6,184 made at least three UDS visits. Finally, 716 participants met two of the five above depression criteria. The majority of participants met depression criteria by self-reported active depression in the last two years (90.6%) in addition to at least one other criteria; a full breakdown of the number of participants who met each respective criterion is presented in S1 Table.

Fig 1

Flowchart of study participant selection process.

All participants included in this study received UDS clinical evaluations at baseline and yearly thereafter.

2.3 Primary outcome and study overview

The primary outcome was first diagnosis of MCI. MCI represents a definable pre-dementia stage of the continuum of cognitive decline. Some persons diagnosed with MCI will not progress to dementia; however, the use of incident MCI as outcome may also yield a higher event rate and more study power within the time period. Diagnosis of MCI was made according to the Petersen criteria if the subject did not have normal cognition and was not clinically demented, but had cognitive complaints not normal for their age, and had largely preserved independence in functional activities[31,32]. MCI is a transitional state characterizing cognitive decline; therefore, if a participant was diagnosed with dementia directly from normal cognition without an intervening MCI diagnosis, then an intermediate MCI stage was assumed to have occurred at the midpoint between the participant’s last normal cognition visit and the visit at which a diagnosis of dementia was made. Such participants were included in the analyses.

2.4 Assessment of antidepressant exposure

Use of medication, including antidepressants, was captured as part of the UDS clinical evaluation, based on clinical interview. Participants were asked to report all prescription medications taken within the two weeks before the current visit.

2.5 Additional clinical measurements

Information on a variety of relevant covariates was collected by the UDS. Demographic characteristics including age, sex, race, and years of education were documented at entry into the study. UDS participants provide a detailed health history (UDS Form A5), completed by a clinician based on the subject’s report, medical records, and observation using the clinician’s best judgement. We defined a list of comorbidities that may confound the relationship among depression, antidepressant use, and cognitive decline, and adjusted in our analyses for history of diabetes, hypertension, hypercholesterolemia, and cardiovascular disease. We adjusted for cigarette smoking based on an initial-visit report of having smoked more than 100 cigarettes in the participant’s lifetime. We also adjusted for apolipoprotein E (APOE) genotype in our models by indicating the presence or absence of at least one e4 allele, a known risk factor for the development of MCI and dementia[33-36].

2.6 Statistical analysis

To investigate the association between antidepressant use and incident MCI, we implemented two types of proportional hazards regression models. The first was a time-dependent Cox proportional hazard model in which antidepressant use was allowed to change over time according to participants’ responses at follow-up visits. This time-dependent approach avoids the potential for immortal time bias. The other type of model used was a fixed-covariate model, in which antidepressant use was categorized and did not change during follow-up. Participants were categorized as either users or non-users of antidepressants in two separate fixed-covariate models: (a) use or non-use of antidepressants at any visit prior to MCI diagnosis visit; (b) use/non-use of antidepressants at baseline, i.e., first visit.

Fixed-covariate model (a) has a major drawback in its susceptibility to immortal time bias: i.e., participants with longer follow-up times will be more likely to eventually use antidepressants and thus be categorized as users. By contrast, fixed-covariate model (b) oversimplifies participants’ real patterns of antidepressant use over time but avoids the possible complications posed by immortal time bias. Immortal time bias is common in studies where participants can encounter the exposure during follow-up. This bias can artificially inflate estimation of the survival time of participants in the exposure group, and therefore could bias results to make antidepressant use appear to reduce the hazard of incident MCI in the absence of a true association. In this model, participants would be categorized as antidepressant users only if the use was reported before the visit where the MCI diagnosis was made. This practice is designed to avoid including participants who began taking antidepressants because of early symptoms associated with MCI.

Finally, we conducted sensitivity analyses to examine the following: (1) How would changing our definition of depression, e.g., from meeting at least two criteria to meeting at least three criteria, affect results? (2) How would simply defining depression as a GDS score of 6 or greater affect the results?

All models described in this study adjusted for age, sex, race, level of education, comorbidity history, smoking history, and the presence of the e4 allele in the APOE genotype. All analyses were conducted using the statistical programming software R (version 3.3.2). For each model, we computed hazards ratios (HRs) and 95% confidence intervals (95% CIs) to assess the risk of developing MCI. The proportional hazards assumption was verified using the Schoenfeld Test [37].

Results

In total, 716 participants met the inclusion criteria for our study (Fig 1). Of those, 464 (64.8%) reported using antidepressants at least once at a visit prior to MCI diagnosis or their final visit (ever-users), and 252 (35.2%) never reported antidepressant use (never-users). Participants had an average length of follow-up of five years in this study. Antidepressant ever-users were slightly younger on average, more likely to be female, less likely to identify as non-white, more likely to have a history of smoking and hypercholesterolemia. Ever-users tended to be consistent users. Of the 464 ever-users, 422 (90.9%) reported antidepressant use during at least half of their UDS visits, and 288 (62.1%) reported using antidepressants at all visits during follow-up. Of the 716 participants, 413 (57.7%) were baseline antidepressant users and 303 (42.3%) were baseline non-users. A summary of baseline characteristics for participants based on their patterns of antidepressant use is given in Table 1.

Table 1

Characteristics of study sample by antidepressant use (N = 716).

	Antidepressant Use
CharacteristicMean (SD) or n (%)	Ever-Users(n = 464)	Never-Users(n = 252)	Baseline Users(n = 413)	Baseline Non-users(n = 303)
Age (years)	71.9 (7.5)	73.4 (7.9)	72.0 (7.4)	73.1 (8.1)
Female	353 (76.1%)	155 (61.5%)	308 (74.6%)	200 (66.0%)
College degree or higher	279 (60.1%)	135 (53.6%)	246 (59.6%)	168 (55.4%)
Non-White Race	53 (11.4%)	44 (17.4%)	41 (9.9%)	56 (18.5%)
1+ APOE-e4 Alleles	126 (27.2%)	61 (24.2%)	114 (27.6%)	73 (24.1%)
Number of visits	5.3 (2.1)	5.2 (2.1)	5.2 (2.1)	5.4 (2.1)
Duration of follow-up (years)	5.1 (2.4)	5.0 (2.3)	5.0 (2.3)	5.2 (2.3)
Smoker (≥ 100 lifetime cigarettes)	235 (50.6%)	108 (42.8%)	211 (51.1%)	132 (43.6%)
Baseline GDS-15a	3.2 (3.1)	4.1 (3.3)	3.0 (3.0)	4.1 (3.6)
Hypertension	237 (51.1%)	135 (53.6%)	215 (52.1%)	157 (51.8%)
Diabetes	61 (13.1%)	40 (15.9%)	52 (12.6%)	49 (16.2%)
Hypercholesterolemia	250 (53.9%)	109 (43.3%)	225 (54.5%)	134 (44.2%)
Cardiovascular Disease	136 (29.3%)	78 (31.0%)	122 (29.5%)	92 (30.4%)

a Geriatric Depression Scale-15

Among the 464 ever-users, 98 (21.2%) eventually developed MCI, compared to 105 of the 252 never-users (41.7%). There were 26 participants (20 ever-users and 6 never-users) who went directly from normal cognition to dementia without an intermediary MCI diagnoses, although for these participants an intermediate MCI stage was assumed to have occurred at the midpoint between their latest normal-cognition visit and their first diagnosis of dementia.

In our primary analysis which utilized a time-dependent Cox proportional hazards model, we did not find an association between antidepressant use and risk of developing MCI (HR = 0.92, 95% CI: 0.70, 1.20; Table 2). However, the association changed when we treated antidepressant use as a fixed covariate. In the first fixed-covariate model, antidepressant ever-users appear to have significantly decreased risk of developing MCI (HR = 0.40; 95% CI: 0.28, 0.56; Table 2). However, when we grouped according to baseline antidepressant use, there was no significant difference between baseline users and non-users (HR = 0.84; 95% CI: 0.61, 1.17; Table 2).

Table 2

Antidepressant use and risk of developing MCI*.

Model and Model Setting	HR for antidepressant exposure	95% CI
Primary analyses a
Time-varying covariate model	0.92	0.70, 1.20
Fixed-covariate model(ever-use vs. never-use)	0.40	0.28, 0.56
Fixed-covariate model(baseline use vs. baseline non-use)	0.84	0.61, 1.17

*Adjusted for age, sex, race, level of education, comorbidity history (diabetes, hypertension, hypercholesterolemia, and cardiovascular disease), smoking history, and the presence of the APOE e4 allele.

a In these primary models, antidepressant use was required to have occurred at least one visit prior to MCI diagnosis or final UDS visit. Participants were considered depressed at entry if they met two of the five criteria for depression.

Finally, we performed two sensitivity analyses to address the impact of depression definition. Inference in these sensitivity analyses were the same as above, however point estimates changed. First, we required that subjects meet at least three depression criteria rather than two. This resulted in a population of 307 subjects who met the revised criteria for depression. A breakdown of the number of participants who met each of the criteria, out of all those who met at least three, is presented in S2 Table. The hazard ratio comparing users to non-users shrank from 0.92 in the original model to 0.69 in this sensitivity analysis (95% CI: 0.47, 1.03; Table 3). In the fixed-covariate model comparing ever-users and never-users, we observed a similar reduction in the estimated hazard ratio from 0.40 in the primary analysis to 0.23 (95% CI: 0.13, 0.38; Table 3). In the fixed-covariate model comparing baseline users and baseline non-users, the estimated hazard ratio shrank from 0.84 in the primary setting to 0.67 in this analysis (95% CI: 0.41, 1.10; Table 3). We then ran an analysis where the definition of depression was based solely on the participants’ baseline GDS score being at least 6 (which reduced the population of analysis-eligible participants to 231). We observed slightly larger hazard ratios in the time-dependent model (HR = 1.18; 95% CI: 0.75, 1.86; Table 3), the fixed-covariate model comparing ever-users to never-users (HR = 0.49; 95% CI: 0.26–0.92; Table 3), and the fixed-covariate model comparing baseline users to baseline non-users (HR = 0.99; 95% CI: 0.54, 1.80; Table 3).

Table 3

Antidepressant use and risk of developing MCI—Sensitivity analyses*.

Model and Model Setting	HR for antidepressant exposure	95% CI
3+ Depression Criteria Required for Entry
Time-varying covariate model	0.69	0.47, 1.03
Fixed-covariate model(ever-use vs. never-use)	0.23	0.13, 0.38
Fixed-covariate model(baseline use vs. baseline non-use)	0.67	0.41, 1.10
GDS ≥ 6 Required for Entry
Time-varying covariate model	1.18	0.75, 1.86
Fixed-covariate model(ever-use vs. never-use)	0.49	0.26, 0.92
Fixed-covariate model(baseline use vs. baseline non-use)	0.99	0.54, 1.80

*Adjusted for age, sex, race, level of education, comorbidity history (diabetes, hypertension, hypercholesterolemia, and cardiovascular disease), smoking history, and the presence of the APOE e4 allele.

Discussion

We did not find a significant association between antidepressant use and risk of incident MCI in our primary analysis which utilized the time-dependent Cox proportional hazards model. We consider this the primary clinical outcome of our work. We also directed our work toward a methodological goal. Previous work has shown that the time-dependent model is the appropriate approach to hazard-based analyses when the exposure of interest can occur during follow-up, most notably because it avoids the potential for immortal time bias that is inherent in these settings[38]. The goal of our implementation of time-fixed models was to provide a practical example of the direction and magnitude of differences which can be obtained based on a modelling choice that invites bias and yet persists in hazards-based analyses [25]. We found that the bias of time-fixed covariate models can be large and varies widely depending on how the time-fixed covariate is defined. When we defined antidepressant use based on ever having reported use, the time-fixed model estimated a significant protective effect on cognitive decline: a hazard ratio of 0.40 as opposed to the hazard ratio of 0.92 estimated by the time-dependent model. Were we to believe the smaller hazard ratio, the overestimation of the effect would enough to misdirect future research. If we define antidepressant use via baseline use we tend to estimate a hazard ratio closer to that of the time-dependent model (0.84 versus 0.92 in our primary comparisons), but the more simplistic model still tends to overestimate the effect. We saw this pattern repeated in sensitivity analyses where the threshold for depression was altered. We hope that this provides an illustration of the impact that naïve modelling choices can have. Future researchers should insist on using time-dependent models to study this association.

We also observed, perhaps predictably, that changing the threshold for depression did impact the results of our analyses. The goal of these sensitivity analyses was to highlight the magnitude of the differences which can be obtained. The comprehensive nature of the UDS allowed us to make these considerations. If we required participants to meet at least three (rather than two) the components of our definition, estimated hazard ratios provided by the three models decreased by 29.2% on average. On the other hand, if we only required that each participant had a baseline GDS score of at least six, hazard ratios increased on average by 22.9%. Even though the statistical significance of the hazard ratios was unchanged in all three models under both sensitivity analysis settings, future studies should be careful to evaluate the severity of depression and be aware of the non-negligible effect that including severely depressed participants or patients with mild depression can have.

We have addressed strategies for researchers who choose to perform (or as a result of available data are driven to) hazards-based analyses. Several prior studies have used hazard-based analyses to examine the association between anti-depressant use and cognitive impairment or dementia[10-12,14,16]. Results are inconclusive; several studies found that antidepressant use is associated with increased risk of developing cognitive impairment and dementia[10-12,14], while another found that use was associated with a decreased risk of developing dementia[16]. None of these studies used time-dependent analyses. Our study extends upon these findings, suggesting no association between anti-depressant use and incident MCI when using robust time-dependent hazard analyses. Goveas et al. conducted a similar analysis in the Women’s Health Initiative Memory Study (WHIMS) but found that antidepressant use was associated with increased risk of MCI [12]. In this study, antidepressant use was assessed according to their current medications and participants were grouped into antidepressant users/non-users at baseline. This is in contrast to our findings when only use anti-depressant at baseline, which may be due to differing study populations (only healthy postmenopausal women were included in WHIMS) or exposure or outcome definitions. Several studies also summarized the relationship between antidepressant use and cognitive decline using longitudinal rates of change (as is in the case in linear mixed model approaches) [19,20] or odds ratios (provided by logistic regression) [17]. Recent studies that have used the linear mixed modeling approach failed to find evidence for a significant association between antidepressant use and cognitive decline[19-21]. In these models, study participants are assigned a cognitive score (usually based on a battery of cognitive tests) and the primary comparison is made between antidepressant users and non-users with respect to the rate of change on this score. These results agree with our finding that the risk of incident MCI faced by antidepressant users and non-users is not significantly different. However, it should be noted that cognitive test batteries may not be sensitive enough to detect true differences in rates of decline between users and non-users, even if they did exist.

On the other hand, it is more difficult to compare our results with those using logistic regression. A previous meta-analysis found that antidepressant users had increased odds of Alzheimer’s disease relative to non-users (OR = 1.65) [9]. Importantly, our work investigated MCI as the outcome of interest and not Alzheimer’s disease. The difference between the conclusion reached here and in the Moraros et al. meta-analysis is large; however, it might be that the cognitive outcome assessed also plays a key role in studies of this association. If that is the case, then it may not necessarily be a contradiction to report that antidepressant use has apparently little impact on risk of incident MCI but plays a larger role in development of dementia or Alzheimer’s disease. Our results can provide some exploratory evidence toward this conclusion. We found that ever-users of antidepressants were less likely to develop MCI than never-users (21.2% of ever-users developed MCI compared to 41.7% of never-users, without adjustment for potential confounders), and we found similar proportions of both groups went on to receive a dementia diagnosis (26.5% of ever-users and 23.8% of never-users). That is, non-users experienced a similar rate of dementia diagnosis despite experiencing twice the rate of MCI diagnoses. This could be an interesting avenue for future research.

Research is also needed in more diverse populations. The UDS is not a nationally representative sample. Most participants in our study were white and more than half had some college education. Since many subjects in the NACC database were volunteers referred by themselves, friends, or family members due to concerns about their memory, observed rates of MCI in this sample may be higher than in the general. Prior studies have also suggested that late-life depression might be accompanied by cognitive decline[5], hence there is a possibility that some of the more depressed participants may already had very mild cognitive impairment at entry into our study despite being categorized as cognitively normal. In the future, participants from a wider array of backgrounds could be evaluated over longer trajectories to address generalizability of our results.

In conclusion, this study illustrates the potential bias in time-fixed models compared to robust methods that account for time-varying exposures. We did not find an association between antidepressant use and risk of incident MCI in our primary analysis which utilized the time-dependent Cox proportional hazards model. The bias of time-fixed covariate models can be large: when we defined anti-depressant use based on ever having reported use, the time-fixed model estimated a significant protective effect of antidepressant use on cognitive decline. However future research is implemented, researchers should be careful to avoid common modelling mistakes and should treat antidepressant use as a time-varying covariate when the data allows. If data are not longitudinal and do not permit such an approach, researchers should carefully consider this limitation. As a general strategy, researchers should recognize the potential for bias when defining time-varying exposures such as depression and antidepressant use and should consider using robust methods and sensitivity analyses to address these difficulties. Future studies with longer follow-up in diverse settings are needed to confirm our finding.

The number and percentage of included participants who met each of the criteria for depression in primary analysis.

(DOCX)

Click here for additional data file.

The number and percentage of participants who met each of the criteria for depression in the first sensitivity analysis.

(DOCX)

Click here for additional data file.

28 Oct 2019

PONE-D-19-27363

Associations between antidepressant use and incident mild cognitive impairment in older adults with depression

PLOS ONE

Dear Dr. Han,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Two reviewers addressed a number of major and minor concerns about your manuscript. Please revise your manuscript carefully.

We would appreciate receiving your revised manuscript by Dec 12 2019 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Kenji Hashimoto, PhD

Academic Editor

PLOS ONE

Journal Requirements:

1. When submitting your revision, we need you to address these additional requirements. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Thank you for stating the following in the Acknowledgments Section of your manuscript:

The NACC database is funded by NIA/NIH Grant U01 AG016976. NACC data are contributed by the NIA-funded ADCs: P30 AG019610 (PI Eric Reiman, MD), P30 AG013846 (PI Neil Kowall, MD), P50 AG008702 (PI Scott Small, MD), P50 AG025688 (PI Allan Levey, MD, PhD), P50 AG047266 (PI Todd Golde, MD, PhD), P30 AG010133 (PI Andrew Saykin, PsyD), P50 AG005146 (PI Marilyn Albert, PhD), P50 AG005134 (PI Bradley Hyman, MD, PhD), P50 AG016574 (PI Ronald Petersen, MD, PhD), P50 AG005138 (PI Mary Sano, PhD), P30 AG008051 (PI Thomas Wisniewski, MD), P30 AG013854 (PI M. Marsel Mesulam, MD), P30 AG008017 (PI Jeffrey Kaye, MD), P30 AG010161 (PI David Bennett, MD), P50 AG047366 (PI Victor Henderson, MD, MS), P30 AG010129 (PI Charles DeCarli, MD), P50 AG016573 (PI Frank LaFerla, PhD), P50 AG005131 (PI James Brewer, MD, PhD), P50 AG023501 (PI Bruce Miller, MD), P30 AG035982 (PI Russell Swerdlow, MD), P30 AG028383 (PI Linda Van Eldik, PhD), P30 AG053760 (PI Henry Paulson, MD, PhD), P30 AG010124 (PI John Trojanowski, MD, PhD), P50 AG005133 (PI Oscar Lopez, MD), P50 AG005142 (PI Helena Chui, MD), P30 AG012300 (PI Roger Rosenberg, MD), P30 AG049638 (PI Suzanne Craft, PhD), P50 AG005136 (PI Thomas Grabowski, MD), P50 AG033514 (PI Sanjay Asthana, MD, FRCP), P50 AG005681 (PI John Morris, MD), P50 AG047270 (PI Stephen Strittmatter, MD, PhD).

We note that you have provided funding information that is not currently declared in your Funding Statement. However, funding information should not appear in the Acknowledgments section or other areas of your manuscript. We will only publish funding information present in the Funding Statement section of the online submission form.

Please remove any funding-related text from the manuscript and let us know how you would like to update your Funding Statement. Currently, your Funding Statement reads as follows:

The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors investigated whether the wide variety of approaches to characterizing antidepressant use, the differing criteria used to define depression, and the multitude of statistical procedures employed by previous studies may contribute to the inconsistent estimates of the association between antidepressant use and risk of Mild Cognitive Impairment (MCI) in older adults with depression. To investigate the association between antidepressant use and incident MCI, they implemented a time dependent Cox proportional hazard model and a fixed covariate model. Treating antidepressant use as a time-varying covariate, they found no significant association with incident MCI. In contrast, when antidepressant use was treated as a fixed covariate, they observed a significant association between having ever used antidepressants and lower risk of MCI. However, in the baseline-use only model, the association was non-significant. Taken together, they concluded that antidepressant use was not significantly associated with incident MCI among cognitively normal persons with depression.

This paper presents evidence that antidepressant use is not significantly associated with incident MCI among cognitively normal persons with depression. Furthermore, this paper demonstrates that methodological approaches play an important role in estimating the association between an exposure and results in epidemiological studies. The findings will be of interest to clinicians, as well as researchers in the field.

I have the following concerns.

1. Methods. P4, line 82. I think the Geriatric Depression Scale (GDS) used here is a short form 15 item GDS, not a 30-item GDS. It would be better to specify it at the first point.

2. Results. P7, line 154. “Participants had an average length of follow-up of five years and spent as many as 11 years in this study.”

The words, “spent as many as 11 years in this study” needs clarification.

3. Results. P8, Table 1. Schreiner et al. (2003) reported that a cut-off score of 6 for the GDS had a sensitivity of 0.97, a specificity of 0.96 for depression. On the other hand, the mean baseline GDS ranges from 3.0 and 4.1 for the subjects, indicating that most of the subjects were below the cut-off point of 6 for the GDS. In the study, subjects were considered depressed if they met at least two of the following five criteria at entry into the study: 1) self-reported active depression in the last two years, 2) depression or dysphoria symptoms as reported by a coparticipant on the NPI-Q, 3) GDS score of at least six, 4) clinically depressed mood based on clinician interview, or 5) a clinical diagnosis of active depression based on current UDS examination and the clinician’s best judgment. I think information on what percentage of subjects meet each criteria should be added to the results.

4. Results. P9, line 187-199. The authors performed two sensitivity analyses to address the impact of depression definition. The number of participants who met each depression definition should be added.

5. Discussion. P13, line 258. “A previous meta-analysis found that antidepressant users had increased odds of Alzheimer’s disease relative to non-users (OR=1.65).”

The authors should cite the relevant paper.

In conclusion, I enjoyed reading this paper. I am grateful that the editor and the authors have given me this kind of opportunity. I think the findings contribute to our understanding of the association between an exposure and an outcome.

Reviewer #2: I enjoyed reading this paper, which was thoughtful and clearly presented, and addressed an important methodological point in the literature. I have a few suggestions for consideration outlined below, divided by section.

Major Revisions

Introduction

Some readers might suggest that time-dependent models are the only acceptable way to analyse data whereby exposures can vary over time. I wonder if this question should be addressed in the Introduction? For instance, what is the purpose of doing a study that compares time-fixed models to a time-dependent model when we know we shouldn’t use time-fixed models? To address this issue, one could state that many studies DO use survival analyses that are sensitive to immortal time bias, and this study helps to demonstrate the potential issues with that approach. Alternatively, are there any downsides to using a time-dependent model in these types of studies? In general, it would be helpful to have a few more details as to the pros and cons of the various analytic techniques to set the stage for the rest of the paper.

Methods

Section 2.1: Consider stating here that many participants were seeking help for subjective cognitive impairment (if this is accurate). I believe that this feature of the study population may be important in interpreting results

In Section 2.3, the authors state that participants with dementia (who were never found to have MCI at UDS visits) were assumed to have passed through an MCI stage. I wasn’t clear whether these participants were included in the analyses?

Results

No suggestions, this section was clearly presented

Discussion

Overall, I find that there is a bit of conflation between the study’s aim of investigating the effects of different methodologies on determining the relationship between antidepressant use and incident MCI and the interpretation of results from a clinical perspective. For instance, much of the Discussion appropriately addresses the effects that various analytic techniques and measurement definitions may have on study outcomes, but in the final paragraph (discussing study strengths), the authors discuss how their primary results rely on a time-dependent proportional hazards model, along with other strengths of the study design. While these factors are true, they are not in keeping with the study aims. Overall, it would be helpful to clarify whether this study is primarily methodological or clinical in nature, and to be consistent throughout the manuscript.

The third and fourth paragraphs require some citations. Specifically, the first and second sentences of the third paragraph in the Discussion require citations regarding which studies use mixed model approaches and odds ratios. Also, in the fourth paragraph, the authors mention a meta-analysis that found an OR of 1.65 for antidepressant users developing Alzheimer’s. I don’t see a citation for this study.

The authors mention that many previous studies addressing similar research questions did not use survival analyses. However, there is at least one that used a comparable survival analysis: the Goveas et al. (2012) study, in which the authors used a Cox proportional hazards model to investigate the same question (though I believe time was fixed; they just examined baseline antidepressant use). This was a different population; specifically cognitively healthy older women who were NOT seeking help for cognitive complaints. It might be interesting to briefly compare the results from this study to the current study, as the methodological differences may explain the discrepancy.

Minor Revisions

Abstract: In the “Discussion” section of the abstract, the first sentence states “…and we suggested that antidepressant use being modeled as a time-varying covariate”. I believe this should state “…and we suggested that antidepressant use should be modeled as a time-varying covariate”.

Section 2.6: The third sentence is “And this time-dependent approach…”. I would suggest removing the “And” and starting the sentence with “This time dependent approach…”.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Kathleen S. Bingham

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

14 Dec 2019

Dear editors and reviewers,

Thank you for your letter and the comments from the referees about our submitted research article titled "Associations between antidepressant use and incident mild cognitive impairment in older adults with depression," (PONE-D-19-27363). Your comments have been valuable to our revision process. We have fully accepted any minor comments on clarity of writing or formatting issues and changes to the text have been made accordingly. The largest revisions occurred in the Introduction and Discussion sections, where we have attempted to clarify the dual purposes of this work: 1) to perform a practical methodological comparison motivated by inconsistencies in previous estimates of this association and the persistence of time-fixed models in the literature even in the presence of time-varying covariates and 2) to highlight our own estimate of the association in question while commenting on the strengths of our modelling approach. Revised portions of the manuscript are marked in red.

Also, if possible at this stage, we want to change the title of our manuscript to a more appropriate one ——"Estimating associations between antidepressant use and incident mild cognitive impairment in older adults with depression".

Funding related text has been removed from the manuscript and is available in the Funding Statement section of the submission form.

Please find our responses to the reviewers’ comments below.

Reviewer 1:

The authors investigated whether the wide variety of approaches to characterizing antidepressant use, the differing criteria used to define depression, and the multitude of statistical procedures employed by previous studies may contribute to the inconsistent estimates of the association between antidepressant use and risk of Mild Cognitive Impairment (MCI) in older adults with depression. To investigate the association between antidepressant use and incident MCI, they implemented a time dependent Cox proportional hazard model and a fixed covariate model. Treating antidepressant use as a time-varying covariate, they found no significant association with incident MCI. In contrast, when antidepressant use was treated as a fixed covariate, they observed a significant association between having ever used antidepressants and lower risk of MCI. However, in the baseline-use only model, the association was non-significant. Taken together, they concluded that antidepressant use was not significantly associated with incident MCI among cognitively normal persons with depression.

This paper presents evidence that antidepressant use is not significantly associated with incident MCI among cognitively normal persons with depression. Furthermore, this paper demonstrates that methodological approaches play an important role in estimating the association between an exposure and results in epidemiological studies. The findings will be of interest to clinicians, as well as researchers in the field.

Thank you for the positive feedback

I have the following concerns.

1. Methods. P4, line 82. I think the Geriatric Depression Scale (GDS) used here is a short form 15 item GDS, not a 30-item GDS. It would be better to specify it at the first point.

Response: The GDS used here is the Geriatric Depression Scale-15 (GDS-15). We now specify this in the manuscript. (P5, line 90)

2. Results. P7, line 154. “Participants had an average length of follow-up of five years and spent as many as 11 years in this study.” The words, “spent as many as 11 years in this study” needs clarification.

Response: We agree that our word choice was confusing. The maximum number of years of follow-up for any participant was 11 years. To avoid confusion, we have altered this sentence to simply report that average length of follow-up: “Participants had an average length of follow-up of five years in this study.” (P8, line 165)

3. Results. P8, Table 1. Schreiner et al. (2003) reported that a cut-off score of 6 for the GDS had a sensitivity of 0.97, a specificity of 0.96 for depression. On the other hand, the mean baseline GDS ranges from 3.0 and 4.1 for the subjects, indicating that most of the subjects were below the cut-off point of 6 for the GDS. In the study, subjects were considered depressed if they met at least two of the following five criteria at entry into the study: 1) self-reported active depression in the last two years, 2) depression or dysphoria symptoms as reported by a coparticipant on the NPI-Q, 3) GDS score of at least six, 4) clinically depressed mood based on clinician interview, or 5) a clinical diagnosis of active depression based on current UDS examination and the clinician’s best judgment. I think information on what percentage of subjects meet each criterion should be added to the results.

Response: We agree this is a useful addition. We have added a set of supplementary tables to the paper to address this issue and the issue below. A breakdown of the percentage of primary-analysis eligible subjects who met each of the criteria for depression is presented in Supplementary Table S1, referenced in the text in section 2.2 (P5, line 95-98). For your reference, we also present that table here. The table below displays the number and percentage (out of the full primary analysis population of 716) of subjects who met each of the criteria. As shown in the table, the number of participants (out of all those who met at least two criteria) who met the self-reported depression criterion was high (649/716; 90.6%). We do recognize that self-reported data may be less reliable, which was a motivation for requiring two depression criteria to be met as well as for our various sensitivity analyses where we address the impact of our composite definition of depression.

S1. The number and percentage of included participants who met each of the criteria for depression in primary analysis.

Criterion Total

(n=716)

Self-reported active depression in the last two years 649 (90.6%)

Clinical diagnosis of active depression based on current UDS examination and the clinician’s best judgement 498 (69.5%)

Depression or dysphoria symptoms as reported by a coparticipant on the NPI-Q 424 (59.2%)

GDS-15 score of at least 6 171 (23.9%)

Clinically depressed mood based on clinician interview 109 (15.2%)

4. Results. P9, line 187-199. The authors performed two sensitivity analyses to address the impact of depression definition. The number of participants who met each depression definition should be added.

Response: Thank you for the suggestion, this information has been presented in Supplementary Table S2, referenced in the text in section 3 (P10, line 197-200), which we also presented here. We felt that the supplementary tables were the best way to present this information because it allows the readers to quickly scan and compare the percentage who met each criterion in the primary analysis and the first sensitivity analysis. The second sensitivity analysis reduced the sample to only those who had a baseline GDS score of 6 or greater. The number of participants eligible for that sensitivity analysis was 231. We also now mention this in the text. Note that this is higher than the 171 participants reported to have met the GDS requirement as part of the primary analysis population, which indicates that there were 60 participants included in the primary analysis who met the GDS criterion but none of the other four criteria.

S2. The number and percentage of participants who met each of the criteria for depression in the first sensitivity analysis.

Criterion Total

(n=307)

Self-reported active depression in the last two years 292 (95.1%)

Clinical diagnosis of active depression based on current UDS examination and the clinician’s best judgement 276 (89.9%)

Depression or dysphoria symptoms as reported by a coparticipant on the NPI-Q 247 (80.5%)

GDS-15 score of at least 6 124 (40.4%)

Clinically depressed mood based on clinician interview 94 (30.6%)

5. Discussion. P13, line 258. “A previous meta-analysis found that antidepressant users had increased odds of Alzheimer’s disease relative to non-users (OR=1.65).”

The authors should cite the relevant paper.

Response: We apologize for this oversite. This paper has now been cited. (P13, line 275)

In conclusion, I enjoyed reading this paper. I am grateful that the editor and the authors have given me this kind of opportunity. I think the findings contribute to our understanding of the association between an exposure and an outcome.

Response: Thank you very much for the helpful comments.

Reviewer 2:

I enjoyed reading this paper, which was thoughtful and clearly presented, and addressed an important methodological point in the literature. I have a few suggestions for consideration outlined below, divided by section.

Major Revisions

Introduction

Some readers might suggest that time-dependent models are the only acceptable way to analyse data whereby exposures can vary over time. I wonder if this question should be addressed in the Introduction? For instance, what is the purpose of doing a study that compares time-fixed models to a time-dependent model when we know we shouldn’t use time-fixed models? To address this issue, one could state that many studies DO use survival analyses that are sensitive to immortal time bias, and this study helps to demonstrate the potential issues with that approach. Alternatively, are there any downsides to using a time-dependent model in these types of studies? In general, it would be helpful to have a few more details as to the pros and cons of the various analytic techniques to set the stage for the rest of the paper.

Response: Thank you for this very important comment. We agree with the suggestion that time-dependent models are the acceptable way of analyzing associations involving time-varying exposures. We have updated the introduction (P3-4，lines 64-70) to clarify our beliefs and highlight that our comparison is intended to show that the difference in hazard ratio estimates provided by time-fixed versus time-dependent models can be quite large in real-world examples. We have also added a citation to the second paragraph（P3，line 63）of the introduction as supporting evidence that the use of time-fixed models remains quite common even when exposures are known to vary over time. We do find that the persistence of these models which are susceptible to immortal time bias motivates a methodological comparison using real-world data.

Methods

Section 2.1: Consider stating here that many participants were seeking help for subjective cognitive impairment (if this is accurate). I believe that this feature of the study population may be important in interpreting results

Response: This is indeed accurate. Many subjects represented in the NACC database are either referred to an Alzheimer’s Disease Center or independently seek out an Alzheimer’s Disease Center due to concerns about their memory. We have added a sentence to section 2.1 to report this. (P4, line 77-79)

In Section 2.3, the authors state that participants with dementia (who were never found to have MCI at UDS visits) were assumed to have passed through an MCI stage. I wasn’t clear whether these participants were included in the analyses?

Response: These participants were included in the analyses. We have added a sentence to the end of section 2.3 to make this clearer (P5, line 112). We also reference this group in section 3: “There were 26 participants (20 ever-users and 6 never-users) who went directly from normal cognition to dementia without an intermediary MCI diagnoses, although for these participants an intermediate MCI stage was assumed to have occurred at the midpoint between their latest normal-cognition visit and their first diagnosis of dementia.” (P9, line 178-181)

Results

No suggestions, this section was clearly presented.

Response: Thank you.

Discussion

Overall, I find that there is a bit of conflation between the study’s aim of investigating the effects of different methodologies on determining the relationship between antidepressant use and incident MCI and the interpretation of results from a clinical perspective. For instance, much of the Discussion appropriately addresses the effects that various analytic techniques and measurement definitions may have on study outcomes, but in the final paragraph (discussing study strengths), the authors discuss how their primary results rely on a time-dependent proportional hazards model, along with other strengths of the study design. While these factors are true, they are not in keeping with the study aims. Overall, it would be helpful to clarify whether this study is primarily methodological or clinical in nature, and to be consistent throughout the manuscript.

Response: Thank you for this comment. As we have mentioned above, the bulk of our revisions have centered around this issue. We consider that this paper has a dual purpose, both methodological and clinical. We have added language in the title, abstract, introduction, and discussion sections to help clarify this point. We focus on methodologic comparisons as a potential explanation for discrepancy in findings on the association between anti-depressant use and cognitive impairment but also think it is valuable to comment on our findings in context of prior literature from a more clinical perspective. Like Reviewer 1 suggests, our aims are for this paper to be relevant to clinicians and researchers alike. We have made revisions throughout the discussion to be consistent with these goals.

The third and fourth paragraphs require some citations. Specifically, the first and second sentences of the third paragraph in the Discussion require citations regarding which studies use mixed model approaches and odds ratios.

Response: Citations for both of these points have been added. (P13, line 262-264)

Also, in the fourth paragraph, the authors mention a meta-analysis that found an OR of 1.65 for antidepressant users developing Alzheimer’s. I don’t see a citation for this study.

Response: Our apologies for this oversite, which was helpfully noticed by both reviewers. We have added the appropriate citation. (P13, line 275）

The authors mention that many previous studies addressing similar research questions did not use survival analyses. However, there is at least one that used a comparable survival analysis: the Goveas et al. (2012) study, in which the authors used a Cox proportional hazards model to investigate the same question (though I believe time was fixed; they just examined baseline antidepressant use). This was a different population; specifically cognitively healthy older women who were NOT seeking help for cognitive complaints. It might be interesting to briefly compare the results from this study to the current study, as the methodological differences may explain the discrepancy.

Response: Thank you for your careful reading. You are correct that Goveas et al. did use survival analysis in their research. We cite this work in the introduction as an example of a study which reported a much higher risk of cognitive impairment among depressed antidepressant users (HR=2.44 in the Goveas study for this group), which is a stark difference from our findings. We now make a brief comparison between our work and Goveas et al. in the third paragraph of the discussion and comment on some possible explanations for this difference. (P13, line 256-262）

Minor Revisions

Abstract: In the “Discussion” section of the abstract, the first sentence states “…and we suggested that antidepressant use being modeled as a time-varying covariate”. I believe this should state “…and we suggested that antidepressant use should be modeled as a time-varying covariate”.

Section 2.6: The third sentence is “And this time-dependent approach…”. I would suggest removing the “And” and starting the sentence with “This time dependent approach…”.

Response: We have revised these sentences. Moreover, we have carefully reviewed the manuscript for any additional instances where the word choice could be improved. If you have additional suggestions, we would be happy to receive them.

Thank you very much for the helpful comments.

Submitted filename: Response to reviewers.docx

Click here for additional data file.

3 Jan 2020

Estimating associations between antidepressant use and incident mild cognitive impairment in older adults with depression.

PONE-D-19-27363R1

Dear Dr. Han,

We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements.

Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication.

Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at https://www.editorialmanager.com/pone/, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

With kind regards,

Kenji Hashimoto, PhD

Section Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: (No Response)

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: (No Response)

Reviewer #2: Thanks for the opportunity to review this paper again. The authors have carefully addressed my comments. I have two minor points. I do not need to see the manuscript again--these points are just for consideration.

Introduction, paragraph 3: A few grammatical suggestions:

"We have two goals for this study; the first is clinical and the second methodological: i) to evaluate the association between antidepressant use and cognitive impairment, and ii) to illuminate the benefits and pitfalls of different methodological approaches and apply relatively robust methods and definitions for inference"

If there is space, consider adding a brief sentence (and maybe an example) in the Intro explaining immortal time bias. Since this paper is also intended for a clinical audience it might be helpful to expand on this a bit more.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Kathleen S. Bingham

10 Jan 2020

PONE-D-19-27363R1

Estimating associations between antidepressant use and incident mild cognitive impairment in older adults with depression.

Dear Dr. Han:

I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

For any other questions or concerns, please email plosone@plos.org.

Thank you for submitting your work to PLOS ONE.

With kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Prof. Kenji Hashimoto

Section Editor

PLOS ONE