Venlafaxine and duloxetine are serotonin norepinephrine reuptake inhibitors (SNRIs) commonly prescribed in older adults for depression, anxiety, and neuropathic pain.
Both medications are metabolized in the liver, and their elimination involves the renal route. The clearance of venlafaxine and duloxetine may be decreased in patients with low kidney function, leading to increased systemic exposure. Venlafaxine and duloxetine use in older adults with low kidney function may increase the risk of developing serious adverse events.
The most common measure to assess kidney function in routine clinical care is the estimated glomerular filtration rate (eGFR), expressed in lab reports as mL/min/1.73 m2. A normal eGFR is \>90 mL/min/1.73 m2, and a value \<45 mL/min/1.73 m2 is considered to be low.
The investigators aim to conduct a population-based cohort study among older adults with low kidney function in Ontario, comparing the risk of serious adverse events among those initiating high-dose SNRIs-venlafaxine (\>37.5-150 mg/day) or duloxetine (\>30-120 mg/day), versus low-dose SNRIs-venlafaxine (37.5 mg/day) or duloxetine (30 mg/day) in the outpatient setting. Secondary analyses will evaluate low-dose venlafaxine (37.5 mg/day) versus low-dose duloxetine (30 mg/day) and high versus lower doses within each drug.
Methods:
In this population-based retrospective cohort study, eligible participants will include older adults (≥66 years) with eGFR \<45 mL/min/1.73 m² (not receiving dialysis or having a history of kidney transplantation) from Ontario who were dispensed a new outpatient prescription for oral venlafaxine or duloxetine. In Ontario, accrual will occur between January 1, 2008, and December 1, 2025.
Based on the prescribed daily dose at treatment initiation, individuals initiating venlafaxine will be categorized into two groups: low-dose (37.5 mg/day) and high-dose (\>37.5-150 mg/day). Similarly, participants initiating duloxetine will be categorized as low-dose (30 mg/day) and high-dose (\>30-120 mg/day) groups.
Propensity score weighting will be used to ensure both groups are well-balanced on a comprehensive set of measured baseline characteristics, including the dose category at treatment initiation. Separate propensity score models will be developed for the secondary analyses.
The primary outcome will be a 30-day composite of all-cause emergency department visit, hospitalization, or mortality.
\*Background\* Venlafaxine and duloxetine are serotonin-norepinephrine reuptake inhibitors commonly prescribed in older adults with low kidney function for major depressive disorder, anxiety, and neuropathic pain, across a range of daily doses. Serious adverse events associated with venlafaxine and duloxetine reported in the literature include hyponatremia, cardiac arrhythmias, hypotension, falls, and central nervous system-related adverse events such as delirium.
Overall, safety data on SNRIs (venlafaxine and duloxetine) in patients with low kidney function are limited, highlighting the need for a population-based study to assess their real-world safety profiles and inform better care.
Understanding the comparative and dose-specific risks of initiating SNRIs in older adults with low kidney function may help inform safer prescribing and improve clinical outcomes in this population.
\*Objective\* Is there a higher 30-day risk of a composite outcome of all-cause emergency department visit, all-cause hospitalization, or all-cause mortality among older adults with low kidney function (an eGFR \<45 mL/min per 1.73 m2 but not receiving dialysis or having a history of kidney transplantation) who initiate high-dose (venlafaxine \>37.5-150 mg/day or duloxetine \>30-120 mg/day) versus low-dose (venlafaxine 37.5 mg/day or duloxetine 30 mg/day) SNRIs in the outpatient setting?
\*Study design and setting\* The investigators propose a population-based retrospective cohort study using linked administrative health data from Ontario. If the Ontario cohort is too small to generate reliable estimates, the investigators will augment the study by performing an analysis using Alberta health administrative data.
Ontario data will be sourced from ICES (Institute for Clinical Evaluative Sciences; ices.on.ca). It offers secure, encrypted individual-level data for Ontario residents, all with universal access to hospital and physician services under a government-funded, single-payer healthcare system. The use of data in this study is authorized under section 45 of Ontario's Personal Health Information Protection Act, which does not require review by a research ethics board. The information needed to analyze the primary outcomes is available across specific databases within the ICES system: data on all-cause hospitalizations are available in the Canadian Institute for Health Information Discharge Abstract Database, emergency department visits are captured in the National Ambulatory Care Reporting System, and mortality is available in the Registered Persons Database.
If the investigators proceed with conducting the study in Alberta and combining the findings with Ontario, the Alberta data will be accessed through the Alberta Kidney Disease Network; this dataset ends in \~ 2021.
The investigators are publicly registering the study protocol and documenting the study description, design, and statistical analysis prior to conducting outcome analyses. The results of this study will be reported adhering to the REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) reporting guidelines.
\*Study Population\* The investigators will include all older adults (≥66 years) with an eGFR \<45 mL/min per 1.73 m2 (not receiving dialysis or having a history of kidney transplantation) who received a new outpatient prescription for oral venlafaxine or duloxetine. The dispensing date of the prescription will serve as the index date. Only the first eligible prescription will be included to ensure unique cohort entry; each individual may enter the cohort only once.
\*Baseline Characteristics\* Health records, census files, hospital discharge records, laboratory data, and physician claims will provide baseline variables, including demographic characteristics (such as age, sex, rurality, neighborhood income quintile, etc.), comorbidities, and medication use. Baseline comorbidities and healthcare utilization will be assessed using 5-year and 1-year look-back periods from the index date. Baseline medication use will be assessed within 120 days prior to the index date.
\*Statistical analysis plan\* Categorical variables will be summarized as frequencies and proportions, while continuous variables will be summarized as means with standard deviations (SDs) or medians with interquartile ranges (IQRs), as appropriate.
Baseline characteristics will be compared between exposure groups using standardized mean differences (SMDs), with an absolute SMD greater than 10% considered indicative of a meaningful imbalance. This approach will be used for the primary comparison between high-dose (venlafaxine \>37.5-150 mg/day and duloxetine \>30-120 mg/day) and low-dose SNRIs (venlafaxine 37.5mg/day and duloxetine 30mg/day ), as well as for secondary comparisons of low-dose venlafaxine (37.5 mg/day) versus low-dose duloxetine (30 mg/day), and between high-dose and low-dose within each drug.
Balancing comparator group: The investigators will use inverse probability of treatment weighting (IPTW) on the propensity score to balance baseline characteristics between the exposure groups, including predictors of venlafaxine and duloxetine use.
Propensity scores will be generated using a multivariable logistic regression model with all baseline characteristics.
Average treatment effect in the treated (ATT) weights will be used, with patients in the low-dose SNRI group will be assigned weights of (propensity score / \[1 - propensity score\]), and patients in the high-dose SNRI group will be assigned a weight of 1. This method produces a weighted pseudo-population in which the distribution of measured baseline characteristics in the low-dose group is similar to that in the high-dose group. Baseline characteristics will be compared between groups using standardized differences in both unweighted and weighted samples.
Regression analysis: To evaluate the primary outcome composite measure of all-cause emergency department visit, all-cause hospitalization, or all-cause mortality, the investigators will apply a modified Poisson regression analysis to estimate the risk ratio (95% confidence intervals) and binomial regression to estimate the risk difference (95% confidence intervals) using the weighted cohort, with the low dose group as the referent.
Secondary analyses:
The investigators will conduct independent testing for all secondary outcomes without adjusting for multiple comparisons. Each outcome will be analyzed and reported independently. In line with best practices, the primary outcome will be presented with its P-value, and the investigators will report all secondary outcomes using point estimates with 95% confidence intervals.
Additional analyses: The investigators plan to conduct seven additional analyses.
1. Effect measure modification (EMM):
To evaluate EMM, the investigators will expand the cohort to all the eGFR levels and categorize them into three groups: eGFR ≥60, 45-\<60, and \<45 mL/min/1.73 m2. Baseline characteristics between the high-dose and low-dose SNRIs groups will be assessed using standardized differences for all renal function categories combined, and within each of the three eGFR categories (≥60, 45-\<60, and \<45 mL/min/1.73 m²).
To further balance baseline characteristics between the high-dose and low-dose SNRI groups, the investigators will apply the IPTW method (described above), based on propensity scores for all eGFR categories combined and within each of the three eGFR categories.
EMM on the absolute scale for the primary composite outcome (30-day composite outcome of all-cause emergency department visit, all-cause hospitalization, or all-cause mortality) will be examined across three baseline eGFR categories.
EMM will be assessed on both the additive and multiplicative scales. For additive interaction, the investigators will use binomial regression with an identity link to estimate risk differences, including an interaction term between high-dose SNRIs and low-dose SNRIs and eGFR strata.
For multiplicative interaction, the investigators will use modified Poisson regression analysis to estimate risk ratios, including an interaction term between venlafaxine and duloxetine and eGFR strata.
2. Our interest in examining the safety of serotonin norepinephrine reuptake inhibitors (venlafaxine and duloxetine) in older adults with low kidney function is derived from high-throughput computing analysis in which the investigators executed 700+ population-based, new-user cohort studies, each comparing 74 acute (30-day) outcomes across strata of kidney function. In this analysis, venlafaxine and duloxetine were compared against non-users, separately; the investigators found signs of harm, prompting us to undertake the current study. High-throughput analysis was run using Ontario data from January 1, 2008, to March 1, 2020. To confirm results, when there is no overlap, the investigators will perform the analysis restricting the cohort after March 1, 2020.
Examine whether the association between high-dose SNRIs versus low-dose SNRIs and the primary outcome is different (modified) in 2 period categories (the period before the end date of the high-throughput computing, and the period after the end date of the high-throughput computing).
The investigators will perform analysis restricted to the non-overlap period, i.e., after March 1, 2020.
3. Compute the hazard ratio for all outcomes within 30 days, illustrating the effect of the intervention on each outcome over time. The investigators expect hazard ratios results similar to risk ratios. Although the difference between hazard ratios and risk ratios may be minimal over a short follow-up period, the hazard ratio offers additional insight by accounting for time-to-event analysis.
4. Compare SNRI users (low-dose and high-dose SNRIs separately) with non-users of SNRIs, separately, to contextualize the magnitude and direction of risk observed in the dose-comparison analysis. This analysis will follow a new-user design and apply propensity score weighting to balance baseline characteristics between study drugs and non-users.
5. The investigators will perform E-value analyses to determine the minimum association strength an unmeasured confounder would need with both the prescription drug and the outcome of interest (while adjusting for measured covariates) to eliminate the observed association.
6. Among patients with low kidney function (eGFR \<45 mL/min/1.73 m²), the investigators will conduct prespecified subgroup analyses to examine whether the association between venlafaxine versus duloxetine users and the 30-day primary composite outcome differs across clinically relevant subgroups.
Subgroups will be defined by baseline eGFR category (30-\<45 and \<30 mL/min/1.73 m²), baseline long-term care residence, anxiety and depression, dementia, and bipolar disorder. Effect measure modification will be assessed on both the additive and multiplicative scales using interaction terms in weighted regression models.
7. The investigators will conduct additional analyses to examine whether the association with the primary 30-day composite outcome differs across dose-specific groups of SNRIs (venlafaxine and duloxetine) among patients with low kidney function. The planned dose-specific comparisons will include: (i) low-dose venlafaxine versus low-dose duloxetine, (ii) low-dose versus high-dose duloxetine, (iii) low-dose versus high-dose venlafaxine.
* Combining Outcome Results from Ontario and Alberta\* This approach will only be used if the investigators proceed with conducting the analysis in Alberta. Privacy-preserving method: The investigators plan to use a privacy-preserving Cox-based approach to estimate risk ratios in multisite studies where individual-level data cannot be shared due to privacy constraints. The proposed method requires only a single transfer of summary-level outputs from each province to the research team. It produces results identical to those from a corresponding log-binomial regression with combined individual-level data. The method was developed by adapting the risk-set table approach for survival outcomes, assuming stratified, province-specific baseline risks, and allowing confounding mitigation strategies, such as propensity score matching or weighting, to be applied individually in each province. Each province will independently calculate these summary tables using its individual-level data. The summary-level risk-set tables generated in Alberta will be shared in a single data transfer to the coordinating site's analyst, who will use them to estimate the combined risk ratios and 95% confidence intervals. This will enable robust, consistent analysis while maintaining data privacy and ensuring regulatory compliance. To comply with privacy regulations for minimizing the chance of identification of any individual, in all manuscripts, numbers of individuals are suppressed in the case of 5 or fewer participants (reported as ≤5). All team members will sign any required data confidentiality and data use agreements.
