Study Design and Setting

We conducted a population-based, retrospective cohort study at the Institute for Clinical Evaluative Sciences (ICES) according to an established protocol approved by the Research Ethics Board at Sunnybrook Health Sciences Centre (Toronto, Canada). Data on adults 66 years of age and older between June 2002 and March 2013 were obtained and analyzed through linked healthcare databases in the province of Ontario. Participant informed consent was not required for this study as all patient information was anonymized and de-identified prior to analysis. The province has about 13.6 million residents, 16% of whom are 65 years or older and have universal coverage for prescription drugs, and 4.7% of whom self-identify as Chinese [18, 20]. The reporting of this study follows guidelines for observational studies (Table A in S1 File) [21]. The date of the first prescription of a study statin served as the index date (also referred to as the cohort entry date or the date of statin initiation).

Data Sources

We ascertained patient characteristics, drug use, covariate information and outcome data using records from five databases. The Ontario Registered Persons Database contains demographic and vital status information for all Ontario residents who have ever been issued a health card. We used the Ontario Drug Benefit database to identify prescription drug use. This database contains highly accurate records (error rate < 1%) for all outpatient prescriptions dispensed to people aged 65 years or older [22]. We identified diagnostic and procedural information on all hospitalizations from the Canadian Institute for Health Information Discharge Abstract Database (CIHI-DAD). We obtained covariate information from the Ontario Health Insurance Plan (OHIP) database, which includes fee-for-service health claims for inpatient and outpatient physician services. Finally, we identified new onset diabetes from the Ontario Diabetes Database (ODD) [23]. We have previously used these databases to research adverse drug events (including outcomes of statin toxicity and health services) [24–30]. We used International Classification of Diseases, 9th revision (ICD-9; pre-2002) and 10th revision (ICD-10; post-2002) codes to assess baseline comorbidities in the three years prior to the index date (Table B in S1 File). The codes we used to ascertain outcomes are detailed in Table C in S1 File, which lists only ICD-10 codes as this was the only coding system in use during the follow-up period.

Classification by Chinese Ethnicity

We identified patients of Chinese and non-Chinese ethnicity using a list of 1,133 unique Chinese surnames prior to anonymization of linked healthcare data. This surname list was developed and validated in Ontario, and has a sensitivity of 80% and a positive predictive value of 92% in classifying individuals who self-identify as Chinese [31]. While people of other Asian ethnicities are included in the non-Chinese group, the proportion is likely small as 77% of the Ontario population does not belong to a visible ethnic minority [32].

Patients

We established a cohort of older adults (i.e. > 65 years) in Ontario, Canada, who initiated one of three study statins (atorvastatin, rosuvastatin, or simvastatin). Other drugs in this class are infrequently used in Ontario, comprising less than 6% of all statin prescriptions during the study period. We excluded patients if they met any of the following criteria: (a) they were in their first year of eligibility for prescription drug coverage (aged 65 years), to avoid incomplete medication records; (b) they received a prescription for any statin (study or non-study) in the 180 days prior to the index date, to confirm new statin use; (c) they were without healthcare eligibility in the 3 years prior to index date (for reasons such as immigration), to ensure sufficient data to assess baseline comorbidities, and (d) they had co-prescriptions for potent CYP3A4 inhibitors (protease inhibitors, chloramphenicol, and antifungals) in the 180 days prior to their index date, to avoid possible statin toxicity from drug-drug interactions [33]. A patient could enter the cohort only once. To select statin users with similar characteristics, Chinese were matched 1:3 to non-Chinese on the following baseline characteristics: age (within two years), sex, index date (same year), chronic kidney disease, coronary artery disease, statin type and dose (high dose defined as ≥10 mg rosuvastatin, ≥20 mg atorvastatin, or ≥40 mg simvastatin; other values were classified as low dose [8]), and the logit of the propensity score for the predicted probability of Chinese ethnicity (within 0.2 standard deviations). The propensity score was derived from a logistic regression model containing 58 appropriately selected baseline variables [34] (Table 1). To measure baseline comorbidity, we used two scoring systems: the Adjusted Clinical Group (ACG) score and the Charlson Comorbidity Index [35–37].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of Chinese and non-Chinese at cohort entry.

https://doi.org/10.1371/journal.pone.0150990.t001

Outcomes

We examined four outcomes associated with statin toxicity as specified in previous studies [7, 10, 24–27, 29, 30]: hospitalization with rhabdomyolysis, incident diabetes, hospitalization with acute kidney injury, and all-cause mortality (the diagnostic codes and their validity are presented in Table C in S1 File). For incident diabetes outcome, we excluded matched sets of patients with evidence of baseline diabetes. We selected acute kidney injury as a potential adverse event of statin toxicity based on a recent report [8]; while completing our study, additional information became available questioning whether acute kidney injury is truly a potential adverse event of a higher serum statin concentration [38, 39]. While liver injury is a historic concern linked to statin therapy, recent studies have not confirmed this association and regulatory bodies no longer identify it as a major concern [40, 41]; thus we decided a priori not to include it as a study outcome. Since up to 25 unique diagnostic codes can be assigned per hospitalization, patients with multiple diagnosis codes were accounted for in the assessment of each hospitalization outcome. However, the overall incidence is underestimated due to a spectrum bias in hospital diagnosis coding, particularly for milder forms of conditions. For example, the incidence of acute kidney injury can be underestimated up to five-fold when assessed by diagnostic codes compared to laboratory values [27].

Statistical Analyses

We compared baseline characteristics of Chinese and non-Chinese using standardized differences [42]. This metric describes differences between group means relative to the pooled standard deviation and is considered a clinically meaningful difference if greater than 10%. We used Cox proportional hazard regression analyses to estimate hazard ratios and 95% confidence intervals for each of our three outcomes (stratifying the models on matched sets). We considered a two-tailed p-value of less than 0.05 to be statistically significant. In the follow-up for these three outcomes, we censored the observation period at the time of death, statin discontinuation, or end of database records (March 31, 2013). In the primary analysis, statin discontinuation was defined by no evidence of a repeat prescription for the same statin within 30 days following the end of the previous prescription day supply. In our data sources, any deaths, dispensed medications, and associated dates are recorded with high accuracy [22, 43]. We repeated the primary analysis in subgroups defined by statin type (atorvastatin, rosuvastatin, or simvastatin) and statin dose (high vs. low, as previously defined). We conducted all statistical analyses using SAS version 9.3 (SAS Institute, Cary, North Carolina).

Baseline Characteristics and Follow-up Observation Period

Cohort selection is presented in Fig A in S1 File. We identified 625,363 new statin users aged 66 or older who were Chinese (n = 20,598) or non-Chinese (n = 604,765). After 1:3 matching we retained 76,132 individuals (19,033 Chinese, 57,099 non-Chinese). The baseline characteristics of patients before and after matching are presented in Table 1. Before matching, those who were Chinese exhibited significant differences in several baseline characterstics and were generally healthier compared to non-Chinese. While Chinese users were more likely to be prescribed lower doses of rosuvastatin than non-Chinese (Table 1), more than half of Chinese were given a starting dose higher than 5 mg per day for each year of the study period, which is contrary to the recommendation from Health Canada and US Food and Drug Administration on rosuvastatin dosing in Asians released in 2005 (Fig B in S1 File). Matching resulted in two groups similar in age, sex, year of statin prescription, statin type, statin dose, baseline healthcare usage, and multiple other baseline characteristics (Table 1).

The mean length of follow-up was 1.1 years and was similar between the groups (1.3 years in Chinese, 1.1 years in non-Chinese; maximum 10.8 years). The total person-years of follow-up was 85,893 (24,259 Chinese, 61,634 non-Chinese). Statin discontinuation during follow-up was lower for Chinese compared to non-Chinese (Table D in S1 File).

Outcomes

Study outcomes are shown in Table 2. Results are expressed in events per 10,000 person-years, as well as in hazard ratios with patients of non-Chinese ethnicity serving as the referent group. There was no significant difference between Chinese vs. non-Chinese in the risk of hospitalization with rhabdomyolysis (hazard ratio [HR] 0.61, 95% confidence interval [CI] 0.28 to 1.34, p = 0.22), incident diabetes (HR 1.02, 95% CI 0.80 to 1.30, p = 0.89), hospitalization with acute kidney injury (HR 0.90, 95% CI 0.72 to 1.13, p = 0.36), or all-cause mortality (HR 0.88, 95% CI 0.74 to 1.05, p = 0.17). We tested the proportionality assumption for all four outcomes using time-dependent covariates in the Cox proportional hazards regression model. The proportionality assumption was not violated for any of the outcomes, with all p-values greater than 0.05. Additionally, we treated death as a competing risk and the results did not differ (Table E in S1 File). The results of subgroup analyses are shown in Fig 1. The risk of rhabdomyolysis was not assessed in subgroups because there were too few events for meaningful analysis. Statin type and dose did not significantly modify the association between Chinese ethnicity and other outcomes. While there was a higher risk of incident diabetes in Chinese patients taking a lower dose of rosuvastatin, the interaction of incident diabetes with higher dose rosuvatatin was not statistically significant (p = 0.07).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Subgroup analysis by statin type and dose for hospitalization with acute kidney.

injury and a diagnosis of diabetes mellitus *†. * High dose defined as ≥10 mg rosuvastatin, ≥20 mg atorvastatin, or ≥40 mg simvastatin; other values classified as low dose. † To comply with privacy regulations numerators of 1 to 5 cannot be reported individually; for this reason we simply present the rate per 10,000 person-years, without additional information on the number of events in each subgroup.

https://doi.org/10.1371/journal.pone.0150990.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Statin-associated adverse outcomes in Chinese and non-Chinese (referent).

https://doi.org/10.1371/journal.pone.0150990.t002

Additional analyses

We conducted several additional analyses after knowledge of the primary results. First, if prescribing physicians titrated the dose of statin differently among Chinese and non-Chinese (to clinical effects such as lipid level), this could abrogate the effect of ethnicity. To assess this we examined the change in statin dose over the course of follow-up and expressed the change standardized to the initial dose ([final dose–initial dose in follow-up] / initial dose × 100). We observed very little change in dose over the follow-up in both Chinese and non-Chinese groups: the median (interquartile range) change in dose for both groups was 0 (0 to 0). We also recorded the frequency and direction of dose changes within the cohort. Doses were stable in 84.0% in Chinese and 86.8% in non-Chinese, and increased in 11.5% of Chinese and 10.4% of non-Chinese.

Second, although we observed no higher rate of statin discontinuation in Chinese versus non-Chinese, it is possible that outpatient blood monitoring differed in the two groups and influenced outcomes. Outpatient blood testing (which is a billable service to the universal provincial healthcare plan) is accurately recorded in our data sources (but values are not available). We observed no significant difference between the two groups in the proportion of those with creatine kinase, serum creatinine, serum glucose, or glycated hemoglobin testing during the follow-up periods (Table F in S1 File).

Third, in the primary analysis statin discontinuation was defined as absence of a prescription renewal within 30 days. In an additional analysis, we lengthened this definition to 100 days (which is the maximum day supply dispensed for a given prescription in Ontario). While the mean follow-up increased to 2.5 years, there remained no higher risk of any of our study outcomes in Chinese compared to non-Chinese (Table G in S1 File).