Association of atypical antipsychotics and mortality for patients hospitalised with pneumonia

Inclusion criteria

We included patients who met the following criteria:

• a) Hospitalisation between October 1, 2001 and September 30, 2012.

• b) 65 years or older on the date of admission.

• c) Discharged with a diagnosis of pneumonia defined as either a primary diagnosis of pneumonia (ICD-9 codes 480.0–483.99 or 485.0–487.0) or a secondary diagnosis of pneumonia with a primary diagnosis of respiratory failure (ICD-9 code 518.81) or sepsis (ICD-9 code 0.38xx).

• d) Had at least one dose of antimicrobial therapy within the first 48 h of admission.

• e) Had at least three or more VA outpatient clinic visits in the year preceding admission.

• f) Received at least one outpatient medication from a VA pharmacy within 90 days prior to admission, thereby ensuring that the patients were receiving medications from VA pharmacies.

We excluded those who received typical antipsychotics from further analyses due to these medications' clear associations with increased mortality. For patients who had multiple pneumonia-related hospitalisations during the study period, we included only their first hospitalisation.

Data sources and definitions

Atypical antipsychotics included risperidone, olanzapine, clozapine, quetiapine, ziprasidone, aripiprazole, amisulpride, paliperidone and sertindole [4, 5, 15].

We controlled for race and ethnicity categories (white, black, and Hispanic), tobacco use (ICD-9 codes 305.1 and V15.82, smoking cessation clinic use, and/or use of medications for the treatment of nicotine dependence such as Zyban, nicotine replacement, or varenicline), alcohol abuse (ICD-9 codes 291, 303, 305.0), and illicit drug use (ICD-9 codes 292, 304, 305, excluding 305.0–0.1). We also used the Charlson–Deyo comorbidity system to identify pre-existing comorbid conditions [16], and the VA priority status for socioeconomic status [17]. Psychiatric conditions, including depression, bipolar disorder, post-traumatic brain disorder (PTSD) and schizophrenia, were identified using the method described by Selim et al. [18, 19].

Potential confounding medications were controlled for by using a count of unique drugs in each of the following classes for outpatient prescriptions filled within 90-days prior to presentation: antidepressants, lithium, statins, beta blockers, calcium channel blockers, oral antidiabetic agents, lipid-lowering agents, other antihypertensive agents, antiarrhythmic agents, inhaled beta agonists, other bronchodilators and theophylline.

Outcomes

The primary outcomes of this study were all-cause mortality within 30- and 90-days of admission. Death was identified using the VA Vital Status file, which has an 98% accuracy in identifying mortality [20].

Statistical analysis

For the primary analyses, we used propensity score matching to balance measured confounders between groups (atypical antipsychotic users versus nonusers). Logistic regression was used to create the propensity score and then nearest-number matching with a calliper of 0.001 with no replacement was performed [21]. We selected candidate variables that we believed would be potentially associated with severity of illness (e.g. intensive care unit admission, use of mechanical ventilation), outcomes (e.g. age, nursing home residence), or with prescription of antipsychotics (e.g. dementia, psychiatric conditions.) Variables included in the propensity score are displayed in table 1. Odds ratios were calculated to determine the association between atypical antipsychotic use and the outcomes using conditional logistic regression.

To analyse the time-to-event for mortality and cardiovascular outcomes by receipt of antipsychotics, we used Kaplan–Meier plots to display the survivor functions and assessed statistical significance using the log rank test.

For secondary analyses, we used generalised linear mixed effect models to examine the association of atypical antipsychotic use on 90-day mortality in the following pre-defined subgroups: dementia, psychiatric conditions (PTSD, schizophrenia, bipolar disorder and depression), pre-existing cardiac conditions (myocardial infarction, heart failure and cardiac arrhythmias), and intensive care unit admission. These models were adjusted for all of the covariates in the propensity score as well as adjusted for the admitting hospital.

Statistical significance was defined as a two-tailed p-value ≤0.05. All analyses were performed using STATA 15 (College Station, TX, USA).