Tolerability of duloxetine in the elderly and in adults: a protocol and preliminary results of a systematic review and individual participant data meta-analysis of randomized placebo-controlled trials

Background. Duloxetine is an antidepressant which benets from a wide range of approval in elderly population while its safety of use in elderly population, compared to younger adults, is not clearly assessed. A comparison of tolerability of duloxetine between elderly and younger adults would help to rule on this issue. Methods and Design. This protocol outlines a systematic review and meta-analysis of individual participant data (IPD) of all double-blind randomized controlled trials comparing the number of serious adverse events among individuals taking duloxetine in comparison to placebo between participants at least 65 years and younger adults in conditions approved by the European Medical Agency (EMA) and the Food Drug Administration (FDA). Secondarily, will be compared the number of any adverse events, clinical ecacy and quality of life between elderly and younger adults under duloxetine, in comparison to placebo. Relevant studies were selected on ClinicalTrials.gov, Clinicaltrialsregister.eu, data sharing platforms, FDA and EMA websites, and from systematic reviews and meta-analyses of duloxetine on PubMed, following Cochrane’s recommendations. Sponsors and authors from eligible studies were invited to share IPD on data sharing platform or directly with our research team. As data cannot be aggregate into a unique database, a two step-approach meta-analysis will be undertaken.

76.8-year-old (Study SAAB) 14 . Skinner et al. (2004) estimated that decrease of oral clearance accelerates at around 52 years of age 15 . Similar results have been reproduced in ulterior pooled analyses of phase III trials 16 with comparable effect sizes (procedure EMEA/H/C/572/II/26 reported a 25 percent-decrease of oral clearance between 29-and 69-year-old individuals). Taken together, these early phase I studies demonstrated an increased systemic exposure to duloxetine in elderly population.
However, the clinical interpretation of pharmacokinetics results remains challenging. Many subgroup analyses of pilot trials did not nd quantitative differences in tolerability between old and younger adults. One plausible explanation lies in the small sample sizes of elderly participants included in these studies 16 15.6% were 65 years of age or over. Although it represents about 1058 subjects, no direct comparison of between-age tolerability across diseases is to be found in the literature. Altogether, it seems that the tolerability of duloxetine in elderly needs clari cation. Therefore, we planned to determine whether use of duloxetine in elderly is associated to a higher risk of serious and non-serious adverse events than in younger adults. Lilly (the sponsor of most duloxetine studies) has a commitment to share their clinical trial data as part of a data sharing policy starting in June 2014. We therefore designed an individual participant meta-analysis of randomized controlled trials of any condition with an FDA or EMA approval of use of duloxetine.

Objectives
Our main objective is to compare the number of serious adverse events (SAE) among individuals taking duloxetine (in comparison to placebo) between participants at least 65 years and younger adults in EMA and FDA approved conditions. Our primary hypothesis is that elderly population is more prone to serious adverse events caused by duloxetine than younger adult population.
Our secondary objectives are to compare the number of any adverse events caused by duloxetine between participants at least 65 years of age and younger adults and to compare both the clinical e cacy of duloxetine on clinical scales and quality of life between elderly individuals and younger adults, separately for each indication. Furthermore, e cacy and tolerability will be compared between young-old (aged between 65 and 75) and old-old participants (aged 75-year-old or more).

Methods And Design
This systematic review and meta-analysis of individual patient data will be conducted of interventional, prospective, double-blind, randomized, placebo-controlled trials with an independent duloxetine arm. According to a protocol registered on PROSPERO (systematic review registration-2019 CRD42019130488), we have conducted a systematic literature review of relevant trials underwent on participants with an age < and ≥ to 65 years old, in population suffering from a condition having an EMA or FDA approval for duloxetine. In this report we describe all eligible studies and present the outcomes that will be extracted for the individual patient data and the methods that will be used for data synthesis following the PRISMA-P recommendations (Additional le 1). The anticipated end date of study is November 2021.

Eligibility criteria
We used the following eligibility criteria: Types of studies: RCTs with both participants ≥ and < to 65 years of age; Types of participants: subjects suffering from a disorder with a known approval for duloxetine by the FDA and the EMA, namely: depression, anxiety, diabetic neuropathic pain, bromyalgia, chronic musculoskeletal pain and stress urinary incontinence; Types of interventions: duloxetine whatever the dosage, the administration frequency, the route of administration; Type of comparator: placebo; Types of outcome measures: report of SAE for each participant under duloxetine and placebo arms and/or non-serious adverse events and/or e cacy and or quality of life; The research was restricted to trials written in English, whatever their publication status (published/unpublished).

Information sources
Searches were conducted in PubMed (to identify individual studies from published systematic reviews and meta-analyses of duloxetine), ClinicalTrials.gov, Clinicaltrialsregister.eu, data sharing platforms (ClinicalStudyDataRequest.com, YODA and Vivli), FDA drug approval packages and on European public assessment report and withdrawn applications from EMA website. The review was performed on studies available on electronic databases from their date of inception to May 31, 2019.

Search strategy
Trials identi cation was systematic with different search strategies depending on the source. First, we search PubMed for all systematic reviews and meta-analyses involving duloxetine in an approved indication. Then, individual trials were identi ed from these systematic reviews and meta-analyses. The search terms were as follows: "(Duloxetine AND Meta-Analysis[ptyp])". In ClinicalTrials.gov, the search was restricted to all interventional studies with adults and older adults, using "duloxetine" as search term. In Clinicaltrialsregister.eu and in data sharing platforms, the search term was "duloxetine" with no lter applied. In FDA website, FDA drug approval packages were downloaded from FDA Approved drug product, entering "duloxetine" as search term. In EMA website, European public assessment reports and withdrawn applications were selected, limited to reports on human with no restriction on the authorization status, using "duloxetine" as search term. The list of included studies, as the list of meta-analyses and of systematic reviews, are reported on additional le 2 and additional le 3, respectively.

Study records
Data management, selection and collection processes Selection and coding of the different study characteristics were performed by two independent reviewers (JCR and AJ) in a blinded manner. A third reviewer (FN) arbitrated in case of disagreement. Studies appearing to duplicate authors, treatment comparisons, sample sizes and outcomes were checked one against another to avoid double-counting and integrating data from several reports on the same study and in contact with the studies sponsors. A data extraction sheet based on the Cochrane Handbook for Systematic Reviews of Interventions guidelines was developed. In case of missing data, the sponsor of the study and/or corresponding authors were contacted.

Collecting IPD
A data sharing request was sent to all sponsors which trials were spontaneously available on data sharing platforms. For the remaining studies, the request was sent to all correspondent authors and, if possible, a research proposal was addressed to each pharmaceutical sponsor on data sharing platforms (for Eli Lilly, Shionogi, P zer and AbbVie trials) or on the sponsor website (for Lundbeck, Takeda and Merck Sharp & Dohme trials). For willing collaborators, the terms of the collaboration will be speci ed in a data transfer agreement, signed by representatives of the data provider and of the recipients (Clinical Investigation Center, Department of Clinical Pharmacology, Rennes University Hospital, France). Collection of IPD is ongoing. Participant characteristics requested are baseline age, gender, intervention arm, duloxetine dose, duration of participation in the study, number of serious adverse events from baseline to endpoint, number of non-serious adverse events from baseline to endpoint, study primary outcome and its values at baseline and at endpoint, type of quality of life scale used and its values at baseline and at endpoint. Data will be accepted in any suitable electronic format. Checks on the data will be made to ensure data are correctly coded, that missing data are correctly identi ed, that extreme values are genuine and to ensure that the data are consistent with published results. Data from all trials will be incorporated into a single database with elds that are consistent across trials.

Data items
For each included study, information was extracted on: Characteristics of the study: year, country, number of arms with duloxetine and placebo, funding, disease; Characteristic of trial participants: mean age (and its standard deviation), gender, number of patients included in analysis, population of analysis used in the identi ed report (intention to treat, per protocol, other); Type of administration and dose; Outcome measures as stated above (including exact de nition of outcome/ e.g. MedDRA or other).

Outcomes and prioritization
The main outcome is the number (count) of SAE for each individual patient. In our study, SAE refers exclusively to any undesirable experience associated with the use of a medical product which results in death, is lifethreatening, requires hospitalization or prolongation of existing hospitalization, results in persistent or signi cant disability, or is a congenital anomaly or birth defect. This outcome has the advantages to be simple of interpretation and clinically relevant for safety estimation, containing a severity criterion from its de nition.
Additionally, will be assessed as secondary outcomes: The number of non-serious adverse events (nsAE). nsAE refers to any untoward medical occurrence in a clinical trial subject administered a medicinal product and which does not have severity criteria for SAE nor necessarily have a causal relationship with this treatment. In a randomized placebo-controlled setting, this outcome provides a global estimate of treatment safety, without the severity feature of SAE.
The e cacy on clinical scale in each indication (i.e. depression, anxiety, pain, urinary incontinence): different scales in the same indication will be standardized by z-scores. This outcome will evaluate the clinical bene t in patients.
The quality of life scores in each indication: the different scales will be standardized by z-score. In studies where multiple quality of life scales were used, a hierarchy was established by selecting scales which both resume health data in one unique total score and which were the most used among the duloxetine trials. If the only available instrument in a trial did not permit to resume data into one score (e.g. SF36), its general health subscale was retained as indicator of quality of life (detailed in Additional le 4, Table S1). Quality of life scores yield a broad and ecological estimate of the well-being state of the patient.

Risk of bias in individual studies
Two researchers (JCR and AJ) assessed each trial for risk of bias independently, addressing randomization, allocation concealment, blinding of assessors and of study participants, completeness of outcome assessment, selective reporting and other potential sources of bias according to the Cochrane Collaboration tool for assessing risk of bias in its current version, RoB2 18 , at the study level. Discrepancies were resolved by consensus.

Data synthesis Main analysis
As it appeared that all data were not directly downloadable together but rather provided remotely on separate interfaces, we chose a two-step procedure 19 to derive incidence rate ratio (for count of binary outcomes) and mean differences (for quantitative outcomes).
The rst step consists in comparing the number (count) of SAE between groups of participants arranged by age using a generalized linear mixed model (with a quasi-Poisson link function) in each individual study. Age (binary variable between "young" participants < 65 years old and "old" participants ≥ 65 years old) and intervention (binary variable between participants under duloxetine to participant under placebo) will be considered as xed effects and their interaction will be explored as the main comparison. Incidence rate ratio as well as their variance will be extracted from the interaction between age and intervention.
In the second step, we will pool the extracted incidence rate ratios using a random effect meta-analysis.
Similarly, number of non-serious adverse events (count) will be analyzed using a two-step approach based on a generalized linear mixed model (with a quasi-Poisson link function). For both SAE and nsAE, we will use metaregressions to explore whether these interactions between age and intervention are different across the different conditions (major depressive disorder, generalized anxiety disorder, diabetic peripheral neuropathic pain, bromyalgia, chronic musculoskeletal pain and stress urinary incontinence).
Summary measures for the analyses of e cacy and quality of life: in each condition, clinical scales (i.e. depression, pain, chronic anxiety) and quality of life scales will be analyzed separately using a similar two-step approach relying on results of a linear model. As various scales might be used across studies, we will use zscores to standardize the different scales and populations.
Missing data will be handled by multiple imputation.

Sensitivity and subgroup analyses
A sensitivity analysis will be performed considering the occurrence of SAE in binary (Yes vs No) with the same two step approach relying on a generalized linear mixed model (with a logistic link function) in each individual study.
A subgroup analysis will be performed applying the same methodology as for adverse events, to explore differences between "young-old" participants (between 65 and 75 years old) and "old-old" participants (>75 years old) in terms of adverse events (serious and non-serious) and of e cacy.

Meta-biases
Missing trials will be explicitly reported with their characteristics and results described. Duplicate publications bias will be limited by the selection and precise identi cation of published and unpublished trials. If one or more trials could not be precisely identi ed by at least an ID number, these trials will be reported in the study results with explained attempts to obtain identi cation.
Countries where studies took place will be systematically assessed in the description studies table, to account for possible location bias. In addition, we clearly outline in our methods that our analysis select only studies written in English.

Con dence in cumulative evidence
We will use GRADE to rate the overall certainty (quality) of evidence that includes the evaluation of risk of bias, inconsistency, indirectness, imprecision and publication factors 20 .

Study selection
Our search identi ed 448 unique studies that were assessed for eligibility (reported in supplementary le). Of these, 77 studies accounting a total of 25 303 randomized participants ful lled the eligibility criteria ( Figure 1).
For searches conducted on PubMed, 48 meta-analyses or reviews were selected from 136 results. Lilly has accepted to share IPD data from 58 studies (Table 1)

5-20
Legend : ITT Risk of bias within studies Figure 2 provides details of the risk of bias assessment from data available online. 35 trials were assessed as being at an overall low risk of bias. Thirty-one trials were assessed as being at unclear risk of bias and 1 at high risk of bias. In this latter study (Perahia2009, HMDI), participants were randomized to duloxetine or placebo after a 32-week exposition to duloxetine, at risk of selecting subjects with a better tolerability to duloxetine.

Conclusion
To our knowledge, this study would represent the rst meta-analysis investigating the safety and utility of duloxetine in elderly population across all conditions approved by European and American regulatory authorities.
Reported results are solely based on online available data. While these suggest an overall low risk of bias across eligible studies, sponsors are yet to be contacted to fully describe characteristic of patients and studies, as to appraise the effective risk of bias of each study. Afterward, a comparison between our current results and results obtained after personal communication with the sponsor can provide an interesting indicator of sponsors' transparency on trials with duloxetine. While 58 studies have been approved for sharing data, these studies all belong to the same sponsor, owner of the duloxetine. The other sponsors are still evaluating our proposal, have not answered to our proposal or have denied to share their data. In addition, IPD data can only be processed in independent platforms forbidding the direct aggregation of data from other data sharing platforms. These hurdles highlight current issues in realizing one-step IPD meta-analysis when multiple sources are concerned, for both industrial and academic sponsors.
The scope of our meta-analysis is intended to bene t to a wide range of medical specialties as an help for the prescribers and as a better care for elderly population. Furthermore, our study exempli es current possibilities and limits of sharing of individual data in meta-research. JCR participated in the development of the study protocol and writing of the manuscript and is responsible for data extraction and eligibility review. CR participated in the study protocol analysis plan and will be responsible for nal analyses. AJ is responsible for data and eligibility review. FN developed the study design and protocol analysis plan and supervises the realization of the study. GR participated in the development of the study protocol and supervises the realization of the study.
All authors read and approved the nal manuscript.