Esophageal and gastric cancer incidence and mortality in alendronate users

Authors

  • Bo Abrahamsen,

    Corresponding author
    1. Department of Medicine F, Gentofte Hospital, Copenhagen, Denmark
    2. OPEN, Institute of Clinical Research, University of Southern Denmark, Odense, Denmark
    • Department of Medicine F, Copenhagen University Hospital Gentofte, DK-2900 Hellerup, Denmark.
    Search for more papers by this author
  • Michael Pazianas,

    1. The Botnar Research Centre, Department of Orthopaedics, Rheumatology & Musculoskeletal Sciences, Oxford University, Oxford, United Kingdom
    Search for more papers by this author
  • Pia Eiken,

    1. Department of Endocrinology & Cardiology, Hillerød Hospital, Hillerød, Denmark
    Search for more papers by this author
  • R Graham G Russell,

    1. The Botnar Research Centre, Department of Orthopaedics, Rheumatology & Musculoskeletal Sciences, Oxford University, Oxford, United Kingdom
    2. NIHR Bone Biomedical Research Unit and Mellanby Centre for Bone Research, University of Sheffield, Sheffield, United Kingdom
    Search for more papers by this author
  • Richard Eastell

    1. NIHR Bone Biomedical Research Unit and Mellanby Centre for Bone Research, University of Sheffield, Sheffield, United Kingdom
    Search for more papers by this author

Abstract

Recent studies have reached conflicting conclusions regarding the risk of esophageal cancer with oral bisphosphonates. Prior studies did not record the number of cancer deaths or endoscopy rates, which could be higher in bisphosphonate users and lead to more cancers being diagnosed at a stage when their esophageal or gastric location could be accurately distinguished. We conducted a register-based, open cohort study using national healthcare data for Denmark. Upper endoscopy frequency, cancer incidence and mortality was examined in 30,606 alendronate users (female, age 50+) and 122,424 matched controls. Primary outcomes were esophageal cancer incidence and death because of esophageal cancer. The analysis showed that alendronate users were more likely to have undergone recent upper endoscopy (4.1 versus 1.7%, p < 0.001). Alendronate users had a lower risk of incident gastric cancer [odds ratio (OR) 0.61; 95% confidence interval (CI): 0.39–0.97) and no increased risk of esophageal cancer (OR 0.71; 95% CI: 0.43–1.19). Risk reductions were greater in users with 10+ prescriptions. The risk of dying of esophageal cancer was significantly reduced in alendronate users after 3 years OR 0.45 (95% CI: 0.22–0.92) but not after 9 years (OR 1.01; 95% CI: 0.52–1.95). An additional comparison with etidronate users revealed no statistically significant difference in outcomes. In conclusion, we found no excess in esophageal cancer deaths or incidence. The early decrease in esophageal cancer rates may relate to the greater use of endoscopy before starting alendronate. Longer term observations also indicated no excess risk of esophageal cancer death and a significantly decreased risk of gastric cancer death. © 2012 American Society for Bone and Mineral Research

Background

Nitrogen containing bisphosphonates (BPs) such as alendronate are very widely used in the prevention of osteoporotic fractures because of proven antifracture efficacy, relatively good tolerability in most patients, and, in many countries, also low cost. Although BPs have antineoplastic and antiangiogenic properties, their capacity to act as esophageal irritants is recognized despite good upper gastrointestinal (GI) tolerability in the phase III trials.1–4 Thus, esophageal and gastric side effects are among the most common reasons for giving up bisphosphonate therapy.5 It is unclear if the esophageal irritation found with BPs in some patients could make the esophagus more susceptible to carcinogens and promote development of dysplasia and cancer. Following a publication in which the FDA presented 23 cases of esophageal cancer in BP users in the United States and 31 from Europe or Japan,6 four epidemiological studies have been conducted.7–10 Two of these studies7, 8 reached conflicting conclusions, although they used the same database in the United Kingdom. Although interesting, these studies did not proceed to examine the influence on cancer deaths. Also, they did not address endoscopy rates, which could be higher in alendronate users and lead to more cancers being diagnosed at an earlier stage and classified more accurately as of esophageal or gastric origin. We examined upper endoscopy frequency, cancer incidence, and cancer mortality in a large, register-based national cohort of postmenopausal women who began alendronate treatment and in background control subjects. We also performed an active comparator cohort study to compare alendronate users with etidronate users.

Methods And Study Population

Data sources

The Danish National Prescription Database (NPD), the Cause of Death register (CoDR), and the National Hospital Discharge Register (“Landspatientregistret,” LPR) were used to examine whether the risk of esophageal or gastric cancer differed between alendronate-treated women and matched control subjects, whether as hospital diagnoses (incidence) or as causes of death (cancer deaths). The LPR is kept by the National Board of Health and maintains a record of all hospitalizations and outpatient appointments including data on discharge diagnoses (International Classification of disease 8 and 10, ICD-8, ICD-10) and diagnostic procedures. All prescriptions at Danish pharmacies are captured in the NPD. Data access was obtained through Statistics Denmark (reference 702538). Ethics committee approval was not needed. The CoDR is maintained by the National Board of Health. Here the causes of death registered in death certificates (immediate, intermediate, underlying, and contributing) are held as ICD-10 codes.

Study cohort

The study was conducted as a restricted cohort study, a design that shares a number of features with the intent to treat design of clinical trials although it is observational and not randomized. We first excluded subjects with a prior hospital diagnosis of any cancer, dating back to the establishment of the LPR in 1977. We then identified the 38,118 new alendronate users, who began treatment between January 1, 1996 and December 31, 2005. Of these, 30,606 were women aged 50+. The year 1995 was used as a run-in year for antiosteoporotic drugs (in order to identify patients who were already on treatment). We assigned to each BP user four unexposed female population control subjects who were born in the same year as the exposed patient and alive on the day that their matched patient filled her first prescription (the index date). The use of strontium ranelate, raloxifene, and bisphosphonates other than etidronate and alendronate was insufficient to allow an active comparator analysis for cancer outcomes. However, we performed an active comparator cohort study using the same methodology and variables, in which 22,609 female alendronate users without prior cancers were age matched with 22,609 female etidronate users.

Statistics

Hospital diagnoses are at present available up to December 31, 2006, so morbidity analyses were run to this point in time while mortality analyses ran to December 31, 2008. Censoring was done on the day of an outcome event, the date of death, or on end of study, whichever occurred first. Women who filled prescriptions for a bisphosphonate at any time during the study period were not allowed to serve as a member of the control cohort, whereas women who received alendronate treatment were analyzed as exposed to alendronate whether they continued treatment, changed treatment to another bisphosphonate, or dropped out of osteoporosis treatment entirely. The alendronate vs etidronate analysis was done in the same way. Adjusted odds ratios (OR) from Cox proportional hazards (PH) analyses were controlled for age, individual Charlson comorbidity index components (as listed in Table 1), the number of comedications, and any prior upper endoscopy or use of proton pump inhibitors (PPIs) in the past year before the index date. Models were prespecified to contain age and Charlson comorbidity entities (Table 1), whereas comedication variables would only be employed if they improved the model beyond the prediction afforded by the comorbid conditions themselves. This led to PPI use being part of the final models. Hormone replacement therapy (HRT), Prednisolone, aceteylsalicylic asid (ASA), and nonsteroidal anti-inflammatory drugs (NSAID) were tested for inclusion in the models but they did not improve the model fit over that given by the comorbid conditions themselves. Where the conditions for Cox PH analysis were not met, we used logistic regression analysis at fixed outcome times (3, 6, and 9 years). For the mortality outcomes, we included as esophageal or gastric cancer deaths all deaths where the ICD-10 codes C15 or C16 appeared on death certificates. There were no death certificates where the cancer diagnoses C15 and C16 appeared together.

Table 1. Baseline Demographics (Alendronate-Treated Cohort and Matched Controls)
 Control subjectsAlendronate users 
N = 122,424N = 30,606
MeanSDMeanSD
Age71.910.071.910.0matched
Charlson index0.41.00.61.2p < 0.001
083.7% 74.7% p < 0.001
1–313.8% 21.3%  
>32.5% 4.1%  
Medications used in last 12 months
 No. of medications5.54.98.46.0p < 0.001
Comorbid conditions
 Myocardial infaction1.8% 2.2% p < 0.001
 Chronic heart failure7.4% 10.3% p < 0.001
 Peripheral vascular disease1.6% 2.4% p < 0.001
 Cerebrovascular disease3.5% 4.0% p < 0.001
 Dementia1.3% 1.1% p < 0.05
 Pulmonary disease3.3% 10.4% p < 0.001
 Collagen disease1.1% 7.1% p < 0.001
 Ulcer disease1.2% 2.4% p < 0.001
 Liver disease, mild0.2% 0.8% p < 0.001
 Liver disease, severe0.0% 0.1% p < 0.001
 Diabetes, w/o complications0.8% 0.7% p = 0.15
 Diabetes, with complications0.8% 0.7% p = 0.15
 Hemiplegia0.1% 0.2% p < 0.001
 Renal disease0.4% 0.3% p = 0.06
Other indicators
 Upper endoscopy last 12 monthsm1.7% 4.1% p < 0.001
 Proton pump inhibitor used9.1% 16.8% p < 0.001

Results

Baseline demographics

Although 75% of women in the alendronate group appeared in fair health with a Charlson comorbidity index of zero, the group had more comorbid conditions and used more comedications than control subjects (Table 1). Alendronate users were twice as likely to have undergone an upper endoscopy in the year before (4.1 vs 1.7%, p < 0.001). We noted a higher frequency of upper endoscopy in study subjects in the year after beginning alendronate treatment; 4.5% versus 2.1% in control (p < 0.001).

Cancer incidence

There were 318 cases of esophageal or gastric cancer, with 44 cases in the alendronate group (0.14%) and 274 (0.22%) in the control group, over a median follow-up time of 3.5 years (range: 1–11 years) to December 31, 2006. Of these 318 persons, 32 (10%) appeared with both esophageal and gastric cancer codes at different contacts. These anatomically ambiguous cancer cases were included in the combined outcome only. Cancer risks were assessed using Cox regression as subjects had unequal length of follow-up and as findings were compatible with the PH model when tested for log-time effect interaction. Cancer-free survival curves are shown in Figure 1. We found a significantly decreased risk of gastric cancer [Table 2; adjusted OR 0.61; 95% confidence interval (CI): 0.39–0.97], a significantly decreased risk of the combined outcome (adjusted OR 0.63; 95% CI 0.45–0.87), and no significant difference in the risk of esophageal cancer (adjusted OR 0.71; 95% CI: 0.43–1.19) between the two study arms. There was no significant difference between weekly and daily alendronate use. Thus, entering a variable to capture any difference in effect between those women who had only used daily alendronate and those who had used weekly alendronate (with or without prior daily treatment), with esophageal cancer as the outcome, revealed an OR of 0.53 (95% CI 0.21–1.34, p = 0.18) nonsignificantly favouring weekly treatment. For gastric cancer, this was 0.79 (0.33–0.92), p = 0.60. For patients who developed esophageal cancer, the median time to event was 2.1 years in the untreated group and 3.0 years in the alendronate group. There was no significant effect modification (ie, interaction) by age, prior upper endoscopy, PPI, or other covariates included in the model. The exception to this was a statistically significant interaction term (p = 0.04) for gastric cancer and pulmonary disease, where alendronate was associated with a significantly reduced gastric cancer risk in patients without pulmonary disease but weakly associated with an increased risk in patients with pulmonary disease (OR 5.45; 95% CI 0.54–55.5, p = 0.15). Although patients who used PPIs were at higher risk of gastric (adjusted OR 3.50; 95% CI 2.26–5.41, p < 0.01) but not esophageal cancer (adjusted OR 1.54; 95% CI 0.84–2.84, p = 0.16), this did not drive or modify the associations found.

Figure 1.

Cox cumulative incidence curves. Incident esophageal (left) and gastric cancer (right) incidence as assessed using diagnoses recorded for hospital contacts (inpatient and outpatient) for the calendar years 1996–2006 (included). Solid lines indicate alendronate-treated women and dotted lines untreated women matched for age. Please refer to Table 2 for crude and adjusted odds ratios and stratification for amount of alendronate used. The patients numbers shown are those for the cancer incidence analysis; please note that the mortality analyses shown in Table 3 have a longer follow-up period as this dataset includes the calendar years 2007–2008.

Table 2. Esophageal and Gastric Cancer Risk
 AlendronateControlNo. of prescriptionsCrude95% CIAdjusted95% CI
N = 30,606N = 122,424OROR
  • Based on outcome data on hospital treatment (outpatient and inpatient diagnoses) ending December 31, 2006. Women aged 50+. Cox proportional hazards analysis, median follow-up time 3.5 years (range: 1–11 years). The adjusted OR is controlled for the number of comedications, for individual comorbidities (Charlson index components in the past 3 years), PPI use and for upper endoscopy (yes/no) done in the past 12 months before initiation of treatment.

  • a

    p < 0.05.

    Note that this analysis compares the cumulative incidence of esophageal and gastric cancer in the alendronate cohort with untreated population control subjects; observations were compatible with the proportional hazards model so outcomes were not shown separately for different lengths of observation as in Table 3.

Esophageal cancer19 (0.06%)99 (0.08%)Any0.75(0.46–1.22)0.71(0.43–1.19)
 17.1/100,00022.9/100,00012.12(0.86–5.22)1.96(0.79–4.89)
   2–90.85(0.34–2.08)0.81(0.32–2.04)
   10+0.53(0.27–1.05)0.51(0.25–1.02)
Gastric cancer22 (0.07%)146 (0.12%)Any0.58(0.37–0.91)a0.61(0.39–0.97)a
 20.0/100,00034.3/100,00011.06(0.39–2.88)1.15(0.42–3.13)
   2–90.91(0.44–1.85)0.99(0.48–2.05)
   10+0.40(0.21–0.75)a0.42(0.22–0.81)a
Combined outcome44 (0.14%)274 (0.22%)Any0.61(0.44–0.84)a0.63(0.45–0.87)a
 40.0/100,00062.9/100,00011.32(0.68–2.57)1.36(0.70–2.67)
   2–90.86(0.50–1.47)0.90(0.52–1.56)
   10+0.42(0.27–0.66)a0.44(0.27–0.69)a

We saw no difference in cancer incidence in the active comparator analysis of alendronate with etidronate as the comparator. For esophageal cancer the OR (aln) was 0.62 (95% CI 0.25–1.53, p = 0.30) and for gastric cancer 0.77 (0.33–1.81), based on 16 versus 22 esophageal cancer cases and 32 versus 20 gastric cancer cases. We found a combined rate of gastric and esophageal cancer of 42 (alendronate) versus 48 (etidronate) cases per 100,000 patient years, that is, lower than the background rate shown in Table 2.

Cancer incidence and cumulative dose

Risk reductions were dose dependent, with the greatest reduction in cancer risk occurring in subjects who filled 10 prescriptions or more. Thus, in patients with 10+ prescriptions the adjusted OR for esophageal cancer was 0.51 (95% CI 0.25–1.02, p = 0.06) and for gastric cancer 0.42 (95% CI 0.22–0.81, p < 0.05), although patients with fewer prescriptions had little or no risk reduction compared with control subjects (Table 2).

Mortality

For the total study (recruited 1996–2005), the cancer mortality analysis had a median follow up of 5.5 years (range: 3–13 years). For esophageal cancer death, we noted significant deviations from the PH model in that the risk was significantly reduced in the first years in the alendronate-treated group of patients but then became similar to those of the control group. Logistic regression analysis confirmed a significantly reduced risk of both esophageal cancer death, of gastric cancer death and of the combined mortality outcome at 3 years (Table 3). There was no statistically significant interaction with age, prior endoscopy, pulmonary disease, or other covariates. At 9 years (recruited 1996–1999), the risk of dying from esophageal cancer was similar to that of controls (OR 0.98; 95% CI: 0.52–1.84. Adjusted OR 1.01; 95% CI: 0.52–1.95). The risk of dying from gastric cancer remained low, but was not significantly reduced compared with controls when adjusted for comorbidities and comedications (OR 0.39; 95% CI: 0.17–0.91; adjusted OR 0.44; 95% CI: 0.19–1.03). The competing risk of dying from other causes, because of greater comorbidity among alendronate-treated subjects, could potentially be the driving force behind the observed reduced risk of developing or dying from specific malignant diseases. However, the all cause mortality at 9 years was lower in alendronate-treated subjects (OR 0.83; 95% CI 0.78–0.89; adjusted OR 0.57; 95% CI 0.53–0.61), corresponding to a 6-months longer mean survival in the alendronate group. In the alendronate-treated population, 47.8% died compared with 52.4% in the control group in the 9-year scenario, whereas 14.8% had died in both groups in the 3-year scenario.

Table 3. Mortality at 3 Years (N = 153,030), 6 Years (N = 60,005), and 9 Years (N = 25,820) After Beginning Alendronate Treatment, Compared With Age-Matched Controls
At 3 yearsAlendronateControlCrude95% CIAdjusted95% CI
N = 30,606N = 122,424OROR
  • Women aged 50+. Logistic regression analysis using outcome information from the cause of death register, ending December 31, 2008. The adjusted OR is controlled for the number of comedications, for individual comorbidities (Charlson index components in the past 3 years), PPI use and for upper endoscopy (yes/no) done in the past 12 months before initiation of treatment.

  • a

    p < 0.05.

    Observations were not compatible with the proportional hazards model so the analysis was done using logistic regression with three different survival windows (3, 6, and 9 years). All numbers are cumulative.

Esophageal cancer9 (0.03%)72 (0.06%))0.50(0.25–1.00)0.45(0.22–0.92)a
Gastric cancer11 (0.04%)98 (0.08%)0.45(0.24–0.84)a0.51(0.27–0.96)a
Combined outcome20 (0.07%)170 (0.14%)0.47(0.30–0.75)a0.48(0.30–0.77)a
All-cause mortality4,541 (14.8%)18,083 (14.8%)1.01(0.97–1.04)0.77(0.74–0.80)a
At 6 yearsN = 12,009N = 48,036    
Esophageal cancer13 (0.11%)73 (0.15%)0.71(0.39–1.28)0.76(0.41–1.41)
Gastric cancer12 (0.10%)90 (0.19%)0.53(0.29–0.97)a0.50(0.26–0.96)a
Combined outcome25 (0.21%)163 (0.34%)0.61(0.40–0.93)a0.62(0.39–0.97)a
All-cause mortality3,855 (32.1%)16,578 (34.5%)0.87(0.83–0.92)a0.58(0.55–0.61)a
At 9 yearsN = 5,164N = 20,656    
Esophageal cancer12 (0.23%)49 (0.24%)0.98(0.52–1.84)1.01(0.52–1.95)
Gastric cancer6 (0.12%)61 (0.30%)0.39(0.17–0.91)a0.44(0.19–1.03)
Combined outcome18 (0.35%)110 (0.53%)0.65(0.40–1.08)0.70(0.42–1.17)
All-cause mortality2,470 (47.8%)10,826 (52.4%)0.83(0.78–0.89)a0.57(0.53–0.61)a

In the smaller active-comparator analysis (alendronate with etidronate as the comparator; Table 4), we found no significant difference in mortality between the treatments. At 3 years, the odds ratio for esophageal cancer death was 0.56 (95% CI 0.18–1.7, p = 0.33) and for gastric cancer death 0.63 (0.28–1.48, p = 0.29), nonsignificantly favoring alendronate. At 9 years, the odds ratio for esophageal cancer death was 2.26 (95% CI 0.99–5.16, p = 0.06). This was based on 12 deaths among 5140 alendronate users with 9+ years of data (22 per 100,000 patient years), with 5 of the 12 having filled only one alendronate prescription. For gastric cancer death the OR was 0.72 (95% CI 0.27–1.89, p = 0.50).

Table 4. Mortality at 3 Years (N = 45,218), 6 Years (N = 31,789), and 9 Years (N = 15,309) After Beginning Alendronate Treatment, Compared With Age-Matched Etidronate Users
At 3 yearsAlendronateEtidronateCrude95% CIAdjusted95% CI
N = 22,609N = 22,609OROR
  • Women aged 50+. Logistic regression analysis using outcome information from the cause of death register, ending December 31, 2008. The adjusted OR is controlled for the number of comedications, for individual comorbidities (Charlson index components in the past 3 years), PPI use and for upper endoscopy (yes/no) done in the past 12 months before initiation of treatment.

  • a

    p < 0.05.

Esophageal cancer5 (0.02%)9 (0.04%)0.56(0.19–1.66)0.33(0.18–1.77)
Gastric cancer9 (0.04%)17 (0.08%)0.53(0.24–1.19)0.64(0.28–1.48)
Combined outcome14 (0.06%)26 (0.11%)0.54(0.28–1.03)0.61(0.31–1.20)
All cause mortality3,431 (15.2%)3,629 (16.1%)0.94(0.89–0.98)a0.88(0.83–0.93)a
At 6 yearsN = 11,866N = 19,923    
Esophageal cancer13 (0.11%)20 (0.10%)1.09(0.54–2.20)1.10(0.54–2.23))
Gastric cancer12 (0.10%)19 (0.0.10%)1.06(0.52–2.19)1.36(0.73–2.51)
Combined outcome25 (0.21%)39 (0.20%)1.08(0.65–1.78)1.08(0.65–1.80)
All cause mortality3,810 (32.1%)6,295 (31.6%)1.02(0.98–1.08)0.97(0.91–1.02)
At 9 yearsN = 5,140N = 10,169    
Esophageal cancer12 (0.23%)12 (0.11%)1.98(0.89–4.41)2.26(0.99–5.16)
Gastric cancer6 (0.12%)15 (0.15%)0.79(0.31–2.04)0.72(0.27–1.89)
Combined outcome18 (0.35%)27 (0.27%)1.32(0.73–2.40)1.36(0.73–2.51)
All cause mortality2,459 (47.8%)4,638 (45.6%)1.09(1.02–1.17)a1.02(0.94–1.11)

Sensitivity analyses

If anatomically ambiguous cases were included among the gastric and esophageal cancer cases, the adjusted risks were very slightly lower both for esophageal and for gastric cancer. Repeating the analyses without patients with a history of upper endoscopy in the last year before beginning treatment did not alter the results.

Discussion

In the present study we demonstrate that postmenopausal women who begin alendronate are not at increased risk of developing esophageal cancer. Further, the risk of dying as a result of esophageal cancer was decreased in the short term, while the risk of dying as a results of gastric cancer was reduced both in the short and the long term. The potential health implications of associations between BPs and cancer are considerable as about 20% of women aged 65+ and above in the United States are treated with BPs.11, 12 Despite a higher frequency of upper endoscopies, the risk of being diagnosed with esophageal cancer was not increased, and the risk of being diagnosed with gastric cancer was strongly reduced. These findings are in support of results recently published by Cardwell and coworkers8, 9 and with findings among Medicare beneficiaries in the United States. We do not find an increased risk of esophageal cancer in bisphosphonate users as was reported by Green et al.7 based on a case–control analysis within the UK General Practice Research Database (GPRD). The present study differs from previous studies by including mortality data and endoscopies; it covers 30,606 bisphosphonate users, 122,424 control subjects followed for up to 13 years, and 118 incident esophageal cancers.

Esophageal cancer is relatively rare in European women, and the age-specific incidence in Denmark is lower than in the United Kingdom or the United States. Some risk factors—especially tobacco and alcohol—are shared between osteoporosis and esophageal cancer13—and could lead to an increased base risk. In accordance with the adverse event profile of nitrogen containing BPs, the frequency of upper endoscopy is increased in patients using oral BPs.14 When this leads to an earlier discovery of esophageal cancers, the incidence rate may increase while the mortality rate will be unchanged or reduced. Some 25% to 40% of esophageal cancers are adenocarcinomas rather than squamous cell carcinomas15 so earlier diagnosis of an adenocarcinoma located just proximal to the gastresophageal junction may enable accurate classification of this as a cancer of the esophagus. Therefore, a higher gastroscopy rate may influence the apparent increased risk of esophageal cancers, leading to an increased risk that is accompanied by a decreased risk of gastric cancer of the same magnitude. Such a relationship between risks was reported in the Green study.7

If ascertainment bias inflates the esophageal cancer rate then cause of death registers become a very important source of corroboration. Both esophageal and gastric cancer have poor survival, with official statistics reporting a 1-year survival of only 26% and 28% in Danish women in the period of time in question, and a 5-year survival of merely 7% and 13%, respectively. Nine-year mortality as assessed here is therefore highly likely to include almost all patients with either of these cancers. In the present study, neither the esophageal cancer mortality, the gastric cancer mortality, nor their combined outcome were increased among alendronate users compared with the background population. We observed a nonsignificant trend towards even lower long term esophageal cancer mortality in etidronate users, but the number of cases was small.

These findings are truly different from the risk increase found by Green et al.7 Thus, Green et al.7 report an odds ratio for esophageal cancer incidence of 1.30 (95% CI 1.02–1.66) for oral bisphosphonates per se. Interestingly, there was no difference in the risks observed with alendronate, risedronate, and etidronate, although the latter is not nitrogen containing. This risk increase contrasts with the odds ratios reported from the two cohort studies, that is, the odds ratio for alendronate of 0.71 (95% CI 0.43–1.19) in the present study and the remarkably similar odds ratio of 0.79 (0.55–1.15) in the GPRD report from Cardwell.8 Further, the risk of esophageal cancer in the Green study was higher if more than 10 BP prescriptions had been filled. This was the subgroup with the lowest risk in the present study. The Green analysis was able to access information about body mass index (BMI), alcohol, and smoking, but so equally was the Cardwell study. Arguably, case–control studies are a more statistically powerful study design when the aim is to identify the risk of one rare outcome. Specifically, given the number of cancer cases observed, the Green study had 90% power to detect a 1.5-fold risk increase, where the present cohort study would need a 1.9-fold risk increase to have the same power. We do not believe this modest difference in study power explains the sizeable difference in mean effects observed between our studies, as a similar difference was found even within the same database of observations, the GPRD.

The influence of competing risks in the two study designs may contribute to the differences. Thus, in the present study and that of Cardwell et al.8 bisphosphonate users were matched to controls of the same age at the time of filling the first prescription. Esophageal cancer events and deaths were then tracked over time, with each subject accruing risk time from initiation of treatment, with alendronate-treated subjects living on average a little longer. Meanwhile, case–control studies identify subjects for study at the date of cancer event. Persons with and without cancer are then assessed for prior exposures. First, in case–control studies, cases and controls are matched so that they are the same age at the time of event, not at the time of first exposure. This leads to the challenge of bisphosphonate-treated patients being of higher average age. In the Green study 44% of bisphosphonate users were aged 80+ compared with 26% of nonusers.7 Second, esophageal cancer is more common in men, so the study design leads to a dominance of men in the analysis, although use is mainly in women. Case–control studies determine relative risks, not event rates. Patients may experience more cancers because of an increased incidence rate, which would be worrisome, or because of a longer life span with an unaltered event rate but a longer life span, which would not. Bisphosphonates have been linked to longevity in mouse models of accelerated aging,16 to increased survival in patients treated for osteoporosis,17, 18 and to an apparent reduction in the risk of breast cancer.19

Esophageal cancer mortality was reduced in the first 3 years after starting treatment but not in the longer term. The decrease should probably be attributed not to a biological effect but to the higher frequency of endoscopy before beginning alendronate, with patients showing endoscopic signs of chonic reflux or early cancer being less likely to be prescribed alendronate. However, for gastric cancer, the reduction in mortality was maintained at 9 years. This raises the possibility that alendronate actively reduces the risk of gastric cancer. There are biologically plausible mechanisms by which bisphosphonates may reduce cancer risk,20–22 although a local antineoplastic effect seems less likely here than in the colon, because of the limited time the drug spends in the stomach.

The major strengths of this study is that it is population based and nationwide, that exposure status is robustly identified while prescriptions that were issued but never used are not included, and that information could be obtained on not only diagnoses and procedures (endoscopies) but also information on deaths, which would be less subject to ascertainment bias.

The dominating position of alendronate on the Danish market precludes a comparison with osteoporosis drugs other than etidronate because of low numbers of users with long-term data. In order to address the concern of confounding by indication, we compared cancer incidence with that found in patients receiving etidronate, which has the same indication as alendronate. Alendronate inhibits esophageal keratinocyte growth23 and has the capacity to cause gastrointestinal irritation and esophageal erosions.24, 25 Etidronate lacks such actions. Because esophageal irritation is associated with Barrett's esophagus and an increased risk of esophageal cancer,26 the comparison of alendronate users with etidronate users is particularly appropriate. There are plausible reasons why nitrogen-containing bisphosphonates such as alendronate might reduce cancer risk. They exert inhibitory effects on the mevalonate pathway, like statins,27 and thereby interfere with intracellular signaling pathways such as those that utilize k-Ras, which are involved in gastrointestinal cancer development.28 In contrast, etidronate does not affect these pathways29, 30 so would not be expected to affect cancer outcomes through these mechanisms.

The only difference between the outcomes in these two groups that was close to statistical significance was a higher number of esophageal cancer deaths in alendronate users over etidronate users with long-term data (average follow-up 10.5 years). The absolute risk was low, with an esophageal cancer mortality rate per 100,000 patient years of 22 in population controls, 22 in alendronate users and 11 in age-matched etidronate users.

This study has limitations. We did not have access to histology reports or individual patient notes for adjudication of diagnoses, nor was information on BMI, or tobacco or alcohol intake available, but estimates were similar to those found by Cardwell8 and which were adjusted for these factors. We included comorbidity entities such as pulmonary disease and liver disease, which would serve in part as a proxy for this information. Thus, the adjusted odds ratios given in the results section take into account the four times higher prevalence of liver disease and three times higher prevalence of chronic obstructive pulmonary disease (COPD) in alendronate users. Patients who begin alendronate may, of course, also be more likely to be aware of general health issues, creating healthy user and healthy adherer bias. Such bias might also contribute to the apparent inverse relationship between adherence and cancer incidence found here. Although postmenopausal women constitute the large majority of bisphosphonate users—85% of the treated population in Denmark11—these drugs are also used by men and occasionally by younger women. Although the difference in esophageal cancer risk associated with bisphosphonate use did not differ significantly by gender in the Cardwell study, a gender difference in effect cannot be completely ruled out as a hazard ratio (HR) of 1.02 (95% CI: 0.68–1.52) was reported in women compared with an HR of 1.22 (95% CI: 0.68–2.20) in men.8

In conclusion, the present nationwide register-based study does not suggest that the risk of developing or dying from esophageal cancer is increased in postmenopausal women who begin alendronate for osteoporosis. Despite a slightly higher frequency of comorbid conditions, women treated with alendronate had a reduced long-term all-cause mortality. The absence of an increased risk of esophageal cancer in an alendronate-treated study population such as the present, which is likely to have higher use of alcohol and tobacco compared with the background population is in itself an important finding given the present concerns that alendronate could cause esophageal cancer. Although the risk of esophageal cancer did not differ significantly from that of population controls or that of patients with osteoporosis treated with the older nonnitrogen containing bisphosphonate, etidronate, the risk of gastric cancer was substantially and dose-dependently reduced compared with the background population.

Disclosures

MP: Consultancy, the Alliance for Better Bone Health and Warner Chilcott. BA: Grant/Research support from Novartis, Nycomed, Amgen, Merck Speakers Bureau with Nycomed, Merck, Eli Lilly. PE: Grant/Research support from Nycomed, Amgen, Novartis Speakers Bureau with Nycomed, Novartis, Eli Lilly and Amgen. RGGR: Research support from Sanofi-Aventis, and Warner Chilcott, Consultant/speaker and legal activities Amgen, GlaxoSmithKline, Roche, Procter and Gamble, Sanofi-Aventis, Novartis, Eli Lilly, and Warner Chilcott. RE: Grant/Research support from Amgen, Novartis, Osteologix, Procter & Gamble, Medtronics, Nestech, Nestle, Fonterra brands, Pfizer, Unipath, Tethys, Inverness Medical, Servier, Ono Pharmaceutical, GlaxoSmithKline, AstraZeneca. Speakers Bureau with Eli Lilly, Takeda, Roche, Amgen, Procter & Gamble, GlaxoSmithKline.

Acknowledgements

The study received grant support from Kaptajnløjtnant Harald Jensen og Hustrus Fond, Denmark. The funders had no influence on the study and did not review the manuscript. Chief physician Jüri Rummessen, MD, DMSci, is thanked for helpful discussions.

Authors' roles: All authors contributed to the study concept, design, and writing of the paper. BA had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. The study received grant support from Kaptajnløjtnant Harald Jensen og Hustrus Fond, Denmark. The funding source had no influence on the study and did not review the manuscript.

Ancillary