Cardiac glycosides use and the risk of lung cancer: a nested case–control study

This study was conducted using the United Kingdom (UK) Clinical Practice Research Datalink (CPRD), formerly known as the General Practice Research Database. The CPRD is the world largest databank on primary care. Since its inception in 1987, it systematically records medical diagnoses and procedures, drug prescriptions issued by general practitioners, patient characteristics (such as body mass index [BMI]), and lifestyle factors (such as smoking and alcohol use). [13] Currently, the CPRD contains data on over 12 million patients registered with more than 650 participating general practices across the UK. Medical diagnoses and procedures are coded using the Read classification, and drugs are coded based on the UK Prescription Pricing Authority Dictionary. Cancer diagnoses, including lung cancer, in the CPRD have been shown to have a high validity [14].

The study protocol was approved by the Independent Scientific Advisory Committee of the CPRD and the Research Ethics Board of the Jewish General Hospital, Montreal, Quebec, Canada.

Within the CPRD population, we identified all patients diagnosed for the first time with HF, AF, AFl and/or SVT, between January 1, 1988 and December 31, 2010, and followed until December 31, 2012. Cohort entry was defined as the date of any of the previously considered diagnoses, whichever appeared first in the patient’s medical record. The cohort was then restricted to patients at least 40 years of age at cohort entry, and those with at least two years of ‘up-to-standard’ medical history in the general practice prior to cohort entry. In order to identify new users of CGs during follow-up, we excluded all patients who previously received these drugs at any time prior to cohort entry. Finally, we excluded all patients previously diagnosed with any cancer (excluding non-melanoma skin cancer) at any time prior to cohort entry to ensure the identification of incident cases of lung cancer during follow-up, and to avoid the inclusion of patients with metastatic disease to the lung from other cancer sites. Patients meeting the study inclusion criteria were then followed until a first-ever diagnosis of lung cancer, death from any cause, end of registration with the general practice, or end of the study period (December 31, 2012), whichever came first.

Within the cohort defined above, we conducted a nested case–control analysis, which produces odds ratios that are unbiased estimators of rate ratios (RRs) (i.e. no need for the rare disease assumption) [15].

Cases consisted of all those newly-diagnosed with lung cancer during follow-up. Up to 10 controls were randomly selected from the case’s risk set (i.e. subset of the cohort still at risk of experiencing the outcome at the time of the case’s event date), after matching on year of birth (±1 year), sex, cohort entry date (±1 year), and duration of follow-up. The date of each case’s lung cancer diagnosis defined the index date, which was also assigned to the matched controls. All controls were alive, not previously diagnosed with lung cancer, and registered with their general practice when matched to a given case. All analyses were restricted to cases and matched controls with at least one year of follow-up in the risk set, which was necessary for latency considerations.

We obtained all prescriptions for CGs received between cohort entry and index date. We excluded exposures initiated in the year immediately prior to index date in order to take into account a latency time window (lag time), and to minimize reverse causality, where initiation or termination of a treatment may have been influenced by early signs or symptoms of lung cancer.

For the primary analysis, exposure to CGs was defined as receiving at least one prescription of digoxin, lanatoside C, digitoxin, or digitalis, between cohort entry and the year prior to index date. For the secondary analysis, we assessed whether there was a dose–response relationship in terms of CG cumulative duration of use and cumulative dose. Therefore, for patients deemed to have ever used CGs, we calculated their cumulative duration of use, defined as the sum of the specified durations of all CGs prescription received between cohort entry and index date. Cumulative dose was computed by multiplying the daily dose of each CG prescription by its specified duration of use and then summing the total quantities received between cohort entry and index date. Since CGs include four different drugs, we used the “defined daily dose” (DDD) equivalence to convert digitalis, digitoxin and lanatoside C in digoxin equivalents doses (the most commonly used CG). Thus, 250 micrograms of digoxin was equivalent to 0.1 milligrams of digitoxin, to 100 milligrams of digitalis, and to 1 milligrams of lanatoside C. Cumulative duration of use and cumulative dose were classified in quartile categories based on the distribution of use in the controls.

Descriptive statistics were used to describe the characteristics of the cohort, cases and matched controls. We used conditional logistic regression to estimate RRs and 95% CIs. In the primary analysis, we assessed whether the use of CGs was associated with an increased risk of lung cancer. In the second analysis, we determined whether there was a dose–response relationship in terms of cumulative duration of use and cumulative dose.

In addition to the matching variables (age, sex, year of cohort entry, and duration of follow-up) on which the logistic regression was conditioned, the models were adjusted for the following potential confounders measured at least one year prior to index date: smoking status, BMI (<18.50 kg/m², 18.50-24.99 kg/m², 25.00-29.99 kg/m², ≥ 30.00 kg/m²), indication of CG use (HF, AF, AFl and/or SVT), excessive alcohol use, history of tobacco-related conditions (chronic obstructive pulmonary disease, ischemic heart disease, and vascular diseases), history of lung diseases (pneumonia, tuberculosis, and history of chronic lung disease), and factors associated with sexual hormonal disorders (hypothalamic, pituitary, testis, ovarian and adrenal gland disorders as well as virilism, hormonal infertility, secondary and primary hormonal deficiency). We also adjusted for drugs potentially associated with lung cancer incidence (also measured at least one year prior to index date), which consisted of statins, aspirin, oral anticoagulants and antiplatelets, non-steroidal anti-inflammatory drugs, anti-hypertensives (diuretics including spironolactone, calcium-channel blockers, angiotensin receptor blockers, angiotensin converting enzyme inhibitors, and beta-blockers), oral bisphosphonates, anti-diabetic drugs (metformin, sulfonylureas, insulins, thiazolidinediones, and other anti-diabetic agents), and amiodarone (which is rather implicated in chronic interstitial pneumonia and commonly prescribed in supraventricular arrhythmia). Variables with missing information were coded with an ‘unknown’ category.

We conducted two sensitivity analyses to assess the robustness of the results. In the first, we varied the lag period prior to index date from one year to six months and two years. The shorter six-month lag period was considered to account for lung cancer’s usual rapid growth. In the second analysis, we additionally adjusted the models for the use of hormone replacement therapy and estrogen-based contraceptives among the female subgroup of cases and matched controls. We also conducted a subgroup analysis, where we assessed whether smoking status, which is the leading risk factor for lung cancer, was an effect modifier of the association between the use of CGs and lung cancer. For this analysis, effect modification was assessed by including interaction terms in the model between CG use and smoking. All analyses were conducted with SAS version 9.3 (SAS Institute, Cary, NC).