Exposure to statins and risk of common cancers: a series of nested case-control studies

We conducted a series of nested case control studies within a cohort of patients registered with practices in the UK contributing to the QRESEARCH database (version 20). The QResearch database (http://​www.​qresearch.​org) is one of the largest general practice databases containing anonymised clinical records for over 11 million patients registered with 574 UK general practices. The information recorded on the database includes patient demographics (year of birth, sex, socio-demographic data derived from UK census 2001), characteristics (height, weight, smoking status), clinical diagnoses, symptoms, and prescribed medications including repeat prescriptions. The database has been validated by comparing birth rates, death rates, consultation rates, prevalence and mortality rates with other data sources, including the General Household Survey and the General Practice Research Database, and has demonstrated good levels of completeness and consistency[23, 24]. Practices were included in the analysis only if they had complete data transmission until at least 1^st July 2008.

We identified an open cohort of patients registered with the study practices during the 10 year study period between 1st Jan 1998 and 1st July 2008. We then used READ codes to select all cases aged between 30 and 100 years with a first record of any cancer in the patients' electronic records occurring during the study period. Each case was linked to 5 controls alive and registered with the practice at the time of diagnosis of the case and matched by age, sex, practice and calendar time. Controls were allocated an index date which was the date on which their matched case was first diagnosed with cancer.

Cases with secondary cancers (READ codes: B56, B57, B58) were excluded. Cases and controls with a diagnosis of any cancer before the index date were excluded. In addition, for breast cancer, we excluded cases and controls with any prior record of mastectomy or prescriptions for tamoxifen since they could be breast cancer cases without a recorded diagnosis in their record. To ensure completeness of exposure data we also excluded temporary residents and patients with fewer than 6 years of medical records before the index date for the main analysis - and fewer than 10 years for the further analysis.

We determined the risks for the most common cancers in the UK[25], comparing these for patients prescribed statins against those not prescribed the drugs. The investigated cancers and corresponding READ codes were: Breast cancer (women, B34), Prostate cancer (men, B46), Lung cancer (B22), Bladder cancer (B49), Haematological malignancies (B6), Gastric cancer (B11), Oesophageal cancer (B10), Colorectal cancer (B13, B14), Pancreatic cancer (B17) and Melanoma (B32). As haematological malignancies cover a range of diseases, possibly differentially affected by statins, we also investigated leukaemia (B63-B6z), lymphoma (B60-B62) and myeloma (B63) separately.

Statin exposure was determined based on all prescriptions for statins until 1 year before the index date (date of diagnosis or equivalent date for controls). The drugs included were atorvastatin, pravastatin, fluvastatin, cerivastatin, rosuvastatin, and simvastatin. Prescriptions in the year before the index date were ignored because including these could lead to results being affected by reverse causality - prescribing in cases in this period might be the result of consultations relating to early cancer symptoms before the recorded diagnosis and this could attenuate any protective effect or exaggerate any harmful effect.

For the main analysis, we considered a 60-month period comprising statin prescriptions for the 13 to 72 months prior to the index date. For the additional analysis, covering a follow-up of 10 years, the period considered was 98 months - for the 13 to 120 months prior to the index date.

Statin use was categorised in a number of ways. We considered a patient as a statin user if they had at least 2 prescriptions in the 60-month period (or the 98-month period for the 10-year analysis). We estimated the cumulative use of statins by extracting the duration of use for every prescription and, for groups of prescriptions with inter-prescription gaps of less than 60 days, we calculated overall course times from the start of the first prescription to the end of the last prescription. We then calculated cumulative use as the sum of all overall course times and for the main analysis categorized cumulative use for each patient as: no use; less than 12 months; 13 to 24 months; 25 to 36 months; 37 to 48 months; 49 to 60 months. A test for trend was performed using the actual number of months of use. For the further analysis, covering a follow-up period of 10 years, the categorisation of the time period for statin use was: no use; less than 12 months, 13 to 24 months, 25 to 48 months, 49 to 72 months, and more than 73 months.

If there were at least 2 prescriptions in the 60-month main study period (or in the 98-month additional study period), we conducted analyses for the following individual statin types: simvastatin, atorvastatin, cerivastatin, fluvastatin, pravastatin and rosuvastatin. For the most common types - simvastatin, atorvastatin and pravastatin - we also examined the effect of cumulative use on cancer risk.

Statin dosage was calculated as median dose across the observation time period, and was categorised as low, medium or high according to statin efficacy[26]. The effect of stopping statin usage on risk of cancer was investigated in the main study only by comparing the last prescription date in the study period with the date one year before the index date and categorising as: no statin use in the 13 to 72 months prior to the index date; still on statins; stopped statins 13 to 24 months before the index date; and stopped statins 25 or more months prior to the index date.

We adjusted for variables which are established cancer risk factors: diabetes[27], rheumatoid arthritis[28], hypertension[29] and body mass index (< 25, 25-29.99, ≥ 30 kg/m²)[30], if recorded at least 1 year before the index date, and for smoking status (non-smoker, ex-smoker, current smoker) and individual Townsend deprivation score (measure of socio-economic status, in fifths), if recorded before the index date. The Townsend score was based on 2001 postcode-related census data, with higher scores indicating greater level of material deprivation and was used because there is a link between deprivation and incidence of some types of cancer[31]. We adjusted for cardiovascular disease as the main reason for statin therapy. For breast cancer we also accounted for any previous benign breast disease (fibrocystic disease, intraductal papilloma, fibroadenoma) and for family history of breast cancer. For colorectal cancer, additional confounders considered were colitis and Crohn's disease.

We also adjusted for use of traditional non steroidal anti-inflammatory drugs, cyclooxygenase-2 inhibitors and aspirin, as several studies have found protective effects for non-steroidal anti-inflammatory drugs and aspirin on various types of cancer[32, 33], in particular on colorectal cancer[33]. We categorised the number of prescriptions for these drugs in the 60-month main study period as: none; 1 to 12; 13 to 24; and 25 or more (adding 25 to 48 and 49 or more for the 98-month additional study period); and adjusted for those categories in assessing cancer risk. We also included in the analyses use of other medications likely to increase the risk of cancer (hormone replacement therapy and oral contraceptive use for breast cancer analysis[34]) if there were at least 2 prescriptions of a drug in the 60-month main study period or 98-month additional study period.

We used conditional logistic regression to estimate odds ratios with 95% confidence intervals for cancer overall and each of the specific cancer sites. As body mass index, smoking status and Townsend deprivation score may be important confounders and have a certain amount of missing data, we used multiple imputation for the missing values[35, 36]. We used the ICE procedure in STATA to obtain 5 imputed datasets and applied Rubin's rules to combine effect estimates and estimate standard errors to allow for uncertainty caused by the missing data. We repeated the imputation procedure for each type of cancer separately.

The initial analysis model determined the unadjusted odds ratios for each cancer associated with statin prescriptions according to: any use of statins in the 60-month study period (at least 2 prescriptions in the 13 to 72 months before the index date); cumulative duration of use; and the median prescribed dose. A multivariate model determined the odds ratio for each cancer associated with statin prescriptions adjusted for the potential confounding effects of variables listed above. For comparison with the analyses using imputed data for smoking status and body mass index, we also ran complete case analyses including only cases and controls with complete data as well as analyses using indicator variables for missing categories of smoking, deprivation and body mass index.

We used all the available data on the QResearch database so did not do a pre-study sample size calculation. According to post-hoc calculation, in order to detect an odds ratio of 0.8 (or 1.2) with 80% power at 1% significance for an exposure that occurs in 15% of controls a sample of 2685 cases (or 3424 cases) would be needed. We checked that we had sufficient power for analysis of the six commoner cancers. STATA v 10 was used for all the analyses. We used a 1% significance level to account for the multiple outcomes.

Table 1 shows baseline characteristics for cases of cancer at any site and their matched controls. Fifty-three percent of the cases were men; the median age at diagnosis was 69 years (interquartile range: 60 to 77). Seventy six percent of cases and 73% of controls had complete data for body mass index, smoking status and Townsend deprivation score.

Cases and controls had similar patterns of co-morbidity except for diabetes (8.1% in cases vs. 7.4% in controls). The difference in proportion of diabetic patients was most marked in pancreatic cancer cases (12.7% vs. 8.3% in controls).

Overall 15.5% of cases and 15.1% of controls had at least 2 statin prescriptions between 13 to 72 months prior to the index date. Most of the statin users (95% of cases and controls) had statin prescriptions for more than a year. Median numbers of scripts for statin users were 19 (interquartile ranges, 9 to 32) for cases and for controls. Median numbers of months on statin were 28 for cases and controls (interquartile ranges, 12 to 50 for cases and 12 to 49 for controls).

The most frequently prescribed statins were simvastatin (9.2% of cases and 9.0% of controls), atorvastatin (6.1% of cases and 5.9% of controls) and pravastatin (1.6% of cases and controls). The other statins were prescribed to less than 1% of the population. Very few atorvastatin users had low dose prescriptions (3 cases and 28 controls) and few pravastatin users had high dose prescriptions (4 cases and 12 controls). Simvastatin dosage was distributed evenly. Long-term statin use was associated with higher dose: in patients prescribed statins for more than 4 years, 42% of cases and 43% controls were on high doses compared with 31% cases and 31% controls on high doses in patients prescribed statins for less than 4 years.

The results of the main analyses, based on patients with at least 6 years of medical records, are shown in Tables 2, 3, 4 and Figure 1. Table 5 shows the odds ratios for each cancer according to cumulative duration of statin use in patients with at least 10 years of medical records.