What this adds

This is a large population-based study to assess the risk of hyponatremia from atypical antipsychotics. The use of an atypical antipsychotic was associated with a modest but statistically significant increase in the 30-day risk of a hospitalization with hyponatremia compared to non-use.

Study design and setting

We used multiple linked health administrative databases to conduct a retrospective population-based cohort study of older adults from June 2003 to March 2012 in Ontario, Canada. There are over two million residents 65 years and older in Ontario who have universal access to hospital care, physician services, and prescription drug coverage [19]. These datasets were linked using unique encoded identifiers and were analyzed at the Institute for Clinical Evaluative Sciences (ICES). We conducted this study according to a pre-specified protocol that was approved by the institutional review board at Sunnybrook Health Sciences Centre, Toronto, Canada. The reporting of this study followed the Strengthening the Reporting of Observational Studies in Epidemiology guidelines (Additional file 1: Table S1) [20].

Data sources

We identified information related to patients, medications, covariates, and outcomes from nine linked databases. For all residents with a valid provincial health card, we obtained demographic information and vital statistics using the Ontario Registered Persons Database. We used the Ontario Drug Benefits Program database to identify exposure to atypical antipsychotics and other medications. This database accurately records prescription claims for outpatients over the age of 65 (error rate of 0.7 %) [21]. We defined covariates using the Ontario Health Insurance Plan database, which contains information on health claims for inpatient and outpatient physician services. We identified diagnostic and procedural information on all hospitalizations and emergency department visits using the Canadian Institute for Health Information’s Discharge Abstract Database and National Ambulatory Care Reporting System database, respectively. Similarly, we used the Ontario Mental Health Reporting System database to obtain mental health information. We obtained antipsychotic prescriber information from the ICES Physician Database. We identified serum sodium measurements using datasets from Gamma-Dynacare Medical Laboratories (a laboratory service provider in Ontario) and Cerner (an electronic medical record system that is used by 12 hospitals in southwestern Ontario). We have used these databases to research adverse drug events and health outcomes in several other studies (including outcomes of hyponatremia and health services) [22–26].

The databases were complete for all variables considered in this study, with the exception of income quintile, rural residence, and prescriber information. We used codes from the International Classification of Diseases, 9th revision (ICD-9) and 10th revision (ICD-10) to ascertain baseline comorbidities in the 5 years prior to cohort entry. Outcomes were identified using only ICD-10 codes as these events would have occurred following implementation of this coding system in 2002. The diagnostic codes used in our study are detailed in Additional file 1: Table S2.

Cohort

For our exposed group, we considered all older adults in Ontario who had evidence of a hospital diagnosis or physician claim for a psychiatric condition (dementia, schizophrenia, bipolar disorder, unipolar depression/anxiety, or Parkinson’s disease) within the previous 5 years and who commenced treatment with risperidone, olanzapine, or quetiapine (users). These are the atypical antipsychotics used most frequently in our region. We defined new use as no prescriptions for any type of antipsychotic drug in the prior 6 months. Patients could only be prescribed one atypical antipsychotic so that we could compare mutually exclusive groups in subgroup analyses. The date of the prescription served as the index date (cohort entry date). We then identified a referent group of older adults with a psychiatric condition, as defined above, from the Ontario population who were not prescribed any kind of antipsychotic (non-users). We randomly assigned an index date to non-users based on the distribution of index dates for the users.

We excluded the following individuals from analyses: (1) patients discharged from hospital in the 2 days prior to their index date to ensure new outpatient antipsychotic use (in the case of the users; or the possibility of a new outpatient antipsychotic prescription in the case of non-users) and (2) patients with end-stage renal disease prior to their index date since serum sodium levels are controlled through dialysis. We excluded all patients with no outpatient medications of any kind dispensed in the 90 days prior to their index date to ensure all were active users of the Ontario Drug Benefits program.

Using a logistic regression model, we derived a propensity score for the predicted probability of commencing treatment with an atypical antipsychotic drug. The propensity score included more than 100 variables that were potentially associated with atypical antipsychotic drug use or hospitalization with hyponatremia (Additional file 1: Table S3) [27]. Then, using greedy matching, we matched 1:1 each atypical antipsychotic drug user to a non-user based on the following characteristics: age (within 2 years); sex; index date (within 1 year); residential status (community-dwelling or long-term care); dementia, schizophrenia, bipolar disorder, unipolar depression/anxiety, Parkinson’s disease, chronic kidney disease, and congestive heart failure; diuretic use; constituency in the catchment area where linked serum sodium data were available; and the logit of the propensity score (within a caliper of ±0.2 standard deviations). A patient could only enter the study once.

Primary outcome

We evaluated the primary outcome of hospitalization with hyponatremia within 30 days following the index date. Hospitalization with any evidence of ICD-10 code E87.1 (hypo-osmolality or hyponatremia) in the patient’s record was included (any one of 25 diagnoses). Based on a validation study conducted in our region, the presence of code E87.1 in a given hospitalization identifies older patients with a median serum sodium value of 123 mmol/L at hospital admission (interquartile range (IQR) 119 to 126 mmol/L), whereas its absence identifies a median value of 138 mmol/L (IQR 136 to 140 mmol/L). The specificity of the code is over 99 %, while sensitivity is 11 % for hyponatremia defined as a serum sodium concentration ≤ 132 mmol/L. The sensitivity of the code increases when hyponatremia is defined by lower serum sodium concentrations [28].

Statistical analyses

We used standardized differences to compare baseline characteristics between user and non-user groups. This metric describes differences between the group means relative to the pooled standard deviation with a value less than 10 % indicating adequate balance [29]. We used conditional logistic regression to estimate odds ratios and 95 % confidence intervals (CIs). We evaluated the association between atypical antipsychotic drug use and a hospitalization with hyponatremia in the following pre-specified subgroups: (1) antipsychotic medication type (risperidone, quetiapine, or olanzapine), (2) antipsychotic dose (higher dose versus normal dose; higher dose defined by a higher than median starting daily dose for the study cohort) (Additional file 1: Table S2), (3) chronic kidney disease, (4) congestive heart failure, and (5) diuretic use. We identified chronic kidney disease and congestive heart failure using separate validated algorithms of hospital diagnostic codes [30, 31]. In the case of antipsychotic type and dose, we defined the subgroup by the characteristic in users with non-users following their matched user. We determined subgroup p values using interaction terms in the logistic regression models.

Additional analyses

We tested the specificity of our findings by evaluating the 30-day risk of a hospitalization with bowel obstruction in the two groups. We expected antipsychotic use would not alter the risk of bowel obstruction and reasoned that a null association with this outcome would enhance causal inference in our hyponatremia analyses. We re-evaluated the outcome of hospitalization with hyponatremia in our existing cohort at a time that preceded the index date by 90 days (a time when no patient would have been prescribed an antipsychotic). After re-applying exclusion criteria at this time, we followed retained matched pairs to re-assess the 30-day risk. In this analysis, a null association would enhance the assertion that the two groups were similar in their baseline risk for hyponatremia in the absence of antipsychotic medication use. We also examined the risk factors associated with hospitalization with hyponatremia separately in users and non-users using multiple logistic regression. The risk factors that we considered were age (per year); sex; chronic kidney disease, congestive heart failure, diabetes, liver disease, cancer, hypothyroidism, and previous hyponatremia; and receipt of a diuretic, antiepileptic, antidepressant, and antineoplastic drugs. We expressed risk in both relative and absolute terms. All odds ratios were approximated as relative risks (appropriate given the incidences observed). We performed all analyses using SAS version 9.4 (SAS Institute, Cary, North Carolina).

Primary outcome

Results from subgroup analyses are presented in Fig. 1. There were too few events among patients with chronic kidney disease for meaningful subgroup analysis, and to minimize any risk of patient re-identification in our data sources, we were not permitted to report results when a group of events were less than or equal to 5. The association between atypical antipsychotic drug use and hospitalization with hyponatremia was not influenced by the atypical antipsychotic drug, dose, congestive heart failure, or diuretic use (all p values for interaction > 0.05).

Additional analyses

The risk of a hospitalization with bowel obstruction was not significantly different between atypical antipsychotic users and non-users (55/58,008 (0.09 %) versus 44/58,008 (0.08 %); relative risk 1.25 (95 % CI 0.84 to 1.86); absolute risk increase 0.02 % (95 % CI −0.01 to 0.05 %)).

In our cohort, older age, cancer, and prior hyponatremia were significant risk factors for future hyponatremia in antipsychotic users (Additional file 1: Table S6).