Higher anti-depressant dose and major adverse outcomes in moderate chronic kidney disease: a retrospective population-based study

Residents of the province of Ontario, Canada have universal access to hospital care and physician services. Those 65 years of age or older, representing approximately 2 million individuals in 2012, also have universal prescription coverage [25]. All health care encounters in Ontario are recorded in linked, de-identified databases at the Institute for Clinical Evaluative Sciences (ICES). We conducted a retrospective, population-based cohort study using six of these healthcare databases. We conducted this study according to a pre-specified protocol that was approved by the research ethics board at Sunnybrook Health Sciences Centre (Toronto, Canada). The reporting of this study follows guidelines for observational studies (detailed in Additional file 1: Figure S1) [26].

We ascertained baseline characteristics, drug use and dose, and outcome data using six linked healthcare databases. Demographic and vital status information on all Ontario residents who have ever been issued a health card is recorded in the Ontario Registered Persons Database (RPDB). Detailed diagnostic and procedural information on all hospital admissions and emergency room visits is recorded in the Canadian Institute for Health Information Discharge Abstract Database (CIHI-DAD) and the National Ambulatory Care Reporting System (NACRS), respectively. Health claims for inpatient and outpatient physician services are recorded in the Ontario Health Insurance Plan database (OHIP). Outpatient prescription drug information including the dispensing date, quantity of pills, dose, and number of days supplied is accurately recorded in the Ontario Drug Benefit Program database (ODB), with an error rate less than 1% [27]. Lastly, the ICES Physician Database (IPDB) contains information on all physicians in Ontario such as sub-specialty, education, location and demographics.

Among a subpopulation of patients with prescriptions filled in Southwestern Ontario, we also obtained baseline serum creatinine values from two linked laboratory datasets: Gamma-Dynacare, a large outpatient provincial laboratory provider and Cerner® (Kansas City, Missouri, USA), an electronic medical record database containing inpatient, outpatient, and emergency department laboratory values for 12 hospitals in Southwestern Ontario [28]. The most recent serum creatinine was obtained in the year prior to the study anti-depressant prescription (median 94 days prior to the prescription). These data sources have been used previously to study drug safety [29‐31]. With the exception of anti-depressant prescriber specialty and income quintile (missing in 13.5% and 0.3% of patients, respectively), the databases were complete for all variables used in this study.

We established a cohort of all older adults in Ontario who had evidence of a new outpatient prescription for a study anti-depressant (defined as no prescriptions for any type of study or non-study anti-depressant in the prior six months) between April 1^st, 2002 and December 31^st 2011 (n = 169,435). The three study anti-depressants were paroxetine, mirtazapine, and venlafaxine. Patients with multiple eligible prescriptions could only enter the cohort once, and the date of anti-depressant initiation served as the patient’s index date (cohort entry date; start of follow-up). We assessed baseline demographic characteristics, co-morbid conditions (5 years prior to index date) and concurrent drug therapy (180 days prior to index date) among all individuals. We excluded the following anti-depressant users from the analysis: those in the first year of eligibility for prescription drug coverage (age 65 years) to avoid incomplete medication records (n = 12,588); those who were discharged from hospital in the two days before their index date to ensure that prescriptions were new outpatient anti-depressant prescriptions (as in Ontario, patients continuing an antidepressant treatment initiated in hospital would have their oral outpatient antidepressant prescription dispensed on the same day or the day after hospital discharge) (n = 3,833); those living in long-term care facilities because some residents chronically experience bouts of confusion (n = 30,360); those with end-stage renal disease since treatments such as dialysis alter anti-depressant pharmacokinetics unpredictably [14] (n = 1,522); and those who received more than one type of anti-depressant on their index date to allow comparison of mutually exclusive exposure groups (n = 3,321). A total of 117,811 patients were included in the final analysis.

We identified individuals with moderate CKD using an algorithm of diagnosis codes validated in our region for older adults [32]. The algorithm identifies a group of patients with a low GFR by the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) formula. It identified patients with a median estimated glomerular filtration rate (eGFR) of 38 mL/min per 1.73 m² (interquartile range 27 to 52), whereas its absence identified patients with a median eGFR of 69 mL/min per 1.73 m² (interquartile range 56 to 82).

To align with recommendations in drug prescribing references, a higher dose of anti-depressant was defined as > 20 mg/day of paroxetine, > 20 mg/day of mirtazapine, or > 37.5 mg/day of venlafaxine (Table 1). A lower dose of anti-depressant was defined as ≤ 20 mg/ day of paroxetine, ≤ 20 mg/day of mirtazapine, or ≤ 37.5 mg/day of venlafaxine (Table 1).

As most reported anti-depressant related delirium occurs within the first few weeks of drug initiation, we followed all individuals for 30 days after first anti-depressant use for two pre-specified outcomes [33]. Our primary outcome was hospitalization with evidence of an urgent head CT scan. We used this as a proxy for the presence of acute central nervous system disturbance (i.e. delirium), as many patients in Ontario undergo such diagnostic imaging when presenting to hospital with acute confusion. Unlike diagnostic codes for acute delirium, the receipt of a head CT scan is well coded in our data sources because they are associated with physician reimbursement [34]. To focus on urgent imaging conducted for acute reasons at the time of hospital admission, we only considered head CT scans performed in the first five days of a hospital admission, or in the emergency department assessment preceding the hospital admission. We expected urgent head CT scans conducted for reasons unrelated to anti-depressant dosing (e.g. stroke, headache) to occur at a similar frequency in higher and lower dose groups, not impacting estimates of difference in risk between the two groups. Our secondary outcome was all-cause mortality which is well coded in our data sources [35].

We compared baseline characteristics between those prescribed higher vs. lower daily doses of study anti-depressants using standardized differences. This metric describes differences between group means relative to the pooled standard deviation and indicates a meaningful difference if greater than 10% [36]. We used multivariable logistic regression analyses (PROC LOGISTIC; SAS Institute, Cary, North Carolina) to estimate adjusted odds ratios and 95% confidence intervals. We adjusted for 15 potential confounders: antidepressant type, age, sex, year of cohort entry, modified Charlson score (a co-morbidity index), and concurrent medication use (anticonvulsants, gabapentin, antipsychotics, barbiturates, benzodiazepines, histamine2-receptor antagonists, dopamine agonists, muscle relaxants, opioids and overactive bladder medications). Outcomes are expressed with patients receiving a lower dose of anti-depressant as the referent group. We then tested for statistical interactions to determine whether the association between anti-depressant drug dose (higher vs. lower) and outcome was modified by the presence of moderate CKD. All odds ratios were interpreted as relative risks (appropriate given the low incidences observed). We conducted all statistical analyses using SAS, version 9.3.

Initiation of a higher vs. lower anti-depressant dose was associated with a lower risk of hospitalization with head CT scan (Table 3; 400/36,651 [1.09%] vs. 1,033/81,160 [1.27%], unadjusted relative risk 0.86 [95% CI 0.76 to 0.96]). After adjusting for 15 potential confounders the association was no longer significant (Table 3; adjusted relative risk 0.90 [95% CI 0.80 to 1.02]). The association was not appreciably different in patients with and without moderate CKD (when the presence of this condition was assessed with diagnosis codes; Figure 1; interaction P value = 0.16). The association was also not appreciably different in patients with and without moderate CKD when examined separately by each of the three anti-depressant drugs (recognizing stratifying the results in this way resulted in smaller sample sizes and wider confidence intervals; Figure 2).

Initiation of a higher vs. lower anti-depressant dose was associated with a lower risk of 30-day all-cause mortality (Table 3; 278/36,651 [0.76%] vs. 791/81,160 [0.97%], absolute risk reduction 0.22% [95% CI 0.10% to 0.33%], unadjusted relative risk 0.78 [95% CI 0.68 to 0.89]). Adjusting for 15 potential confounders had no appreciable impact on this observed association (adjusted relative risk 0.82 [95% CI 0.71 to 0.95; Table 3). The association was not appreciably different in patients with and without moderate CKD (when the presence of this condition was assessed with diagnosis codes; Figure 1; interaction P value = 0.96). The association was also not appreciably different in patients with and without moderate CKD when examined separately by each of the three anti-depressant drugs (Figure 2).

We repeated our analysis in a subpopulation of older adults with baseline serum creatinine values (n = 24,641) comparing a higher vs. lower anti-depressant dose and each of our two outcomes. In this analysis the clinically important threshold used to define moderate CKD was an eGFR < 45 mL/min per 1.73 m² (as higher eGFR thresholds (45 to 60 mL/min per 1.73 m²) may not identify substantial CKD in the elderly, and choosing a lower eGFR threshold (<30 mL/min per 1.73 m²) meant less patients (n = 587 for this analysis)). Results are presented in Figure 3. Point estimates of the relative risk of a higher vs. lower anti-depressant dose and outcomes of head CT and all-cause mortality were all below a value of 1 in patients with and without a baseline eGFR value < 45 mL/min per 1.73 m². The association was not appreciably different in those with and without moderate CKD (Figure 3; interaction P value = 0.73 and 0.19 for each outcome, respectively).

In the primary cohort we considered the outcome of hospitalization using diagnostic codes for delirium (recognizing the coding for this outcome is insensitive and underestimates events but was expected not to operate differently in the two anti-depressant dose groups). Initiation of a higher vs. lower anti-depressant dose was not associated with a difference in risk of hospitalization with delirium (43/36,651 [0.16%] vs. 126/81,160 [0.12%], unadjusted relative risk 0.76 [95% CI 0.53 to 1.07]). The association was not different in patients with and without moderate CKD (interaction P value = 0.75).