We did a population based, nested case-control study of Ontario residents aged 66 years or older treated with spironolactone between 1 April 1992 and 1 March 2010. We determined prescription drug use by using the Ontario Drug Benefit Database, which identifies prescriptions dispensed to Ontario residents aged 65 years or older. Hospital admission data came from the Canadian Institute for Health Information’s Discharge Abstract Database and demographic information from the Registered Persons Database. The Discharge Abstract Database contains clinical information on all admissions, discharges, and same day surgeries from all hospitals in Ontario. Trained health information professionals abstract patients’ charts by using standard diagnosis and procedure codes. We used the Ontario Health Insurance Plan database to identify claims for physicians’ services and the Ontario Diabetes Database for information on diagnoses of diabetes.15
These databases were linked in an anonymous fashion by using encrypted health card numbers and are regularly used to study drug safety, including the consequences of drug interactions.5 16 17 18
For each patient, we identified a period of continuous use of spironolactone beginning with the first prescription for spironolactone after the patient’s 66th birthday. We excluded the first year of eligibility for coverage of prescription drugs (age 65) to avoid incomplete drug records. Observation ended with the first occurrence of a hospital admission for hyperkalaemia, death, the end of the study period (31 March 2010), or cessation of spironolactone treatment, defined as a lapse of more than 180 days between prescriptions. In the event of such a lapse, we extended the observation period 180 days from the date of the last prescription to identify outcomes that may have precipitated cessation of treatment.
Within the cohort of continuous users of spironolactone, we defined cases as those admitted to hospital with a diagnosis of hyperkalaemia (international classifications of diseases, 9th edition, code 276.7, and 10th edition, code E87.5) within 14 days of receiving a prescription for one of four study antibiotics: trimethoprim-sulfamethoxazole, norfloxacin, nitrofurantoin, or amoxicillin. We did not include prescriptions for trimethoprim monotherapy, because the drug is invariably used in combination with sulfamethoxazole in Canada. We included only patients who had hyperkalaemia at the time of admission, rather than those who developed hyperkalaemia during the course of a hospital admission. We restricted our analyses to antibiotics that are primarily used to treat urinary tract infections to avoid the potential confounding effects of other systemic infections. The date of hospital admission served as the index date for all analyses, and we considered only the first instance of hospital admission for hyperkalaemia for patients with more than one such admission during the study period. We excluded patients if they received prescriptions for multiple antibiotics or any non-study antibiotics in the 30 days before the index date.
From within the cohort of patients receiving spironolactone, we selected up to four controls for each case by using incidence density sampling.19
We required controls to have had no hospital admission for hyperkalaemia before the index date and to have received one of the study antibiotics within 14 days before the index date. Consequently, all cases and controls were older patients receiving spironolactone who had also received treatment with one of the study antibiotics. We matched controls and cases on age at the index date (plus or minus one year), sex, presence or absence of diabetes (based on review of the validated Ontario Diabetes Database15
), and presence or absence of chronic kidney disease, determined from physicians’ claims, hospital admission records, and receipt of dialysis in the year before the index date. Each patient could serve only once as a control. When fewer than four control patients were available for each case, we analysed only those controls and maintained the matching process. We excluded any cases that could not be matched to at least one control.
We calculated descriptive statistics for baseline demographic and clinical characteristics of cases and controls, as well as standardised differences to test for differences between the two groups. Standardised differences of less than 0.1 indicate a good balance between the cases and controls for a given covariate.20
We used conditional logistic regression to estimate the odds ratio and 95% confidence intervals for the association between hyperkalaemia related hospital admission and receipt of a prescription for trimethoprim-sulfamethoxazole in the preceding 14 days, using amoxicillin treated patients as the reference group. We selected amoxicillin as the reference antibiotic because, with rare exceptions, it should not cause hyperkalaemia.21
To test the specificity of our findings, we also examined the association between hyperkalaemia and prescription of nitrofurantoin or norfloxacin, antibiotics commonly used to treat community acquired urinary tract infections. We hypothesised that these drugs would not be associated with an increased risk of hyperkalaemia relative to amoxicillin. We used multivariable conditional logistic regression to adjust for medical conditions and classes of prescription drugs that may influence the risk of hyperkalaemia (see web appendix).22 23 24 25
To ascertain whether a dose-response relation existed between trimethoprim and hyperkalaemia, we repeated our primary analysis, stratifying exposure to trimethoprim-sulfamethoxazole according to prescription for single strength (400 mg/80 mg) or double strength (800 mg/160 mg) tablets. Finally, we determined the population attributable fraction, defined in this study as the fraction of all cases (exposed and unexposed) that would not have occurred if exposure to trimethoprim-sulfamethoxazole had been avoided.26 27
We used SAS version 9.2 for all analyses.