Case-control analysis within a UK primary care cohort: oral bisphosphonates and risk of cancer of oesophagus, stomach, and colorectum

Why do the study?

Oral bisphosphonates are commonly used in the treatment of osteoporosis, Paget’s disease, myeloma, and metastatic bone disease. When drugs are widely used, it is important that side effects and potential complications of prolonged use are reported in medical literature to guide the need for further research and inform prescribing practice. Bisphosphonates are reasonably well tolerated, but gastrointestinal side effects ranging from mild gastric discomfort through to ulcer disease have been documented. In 2009, the possibility of a link between the use of bisphosphonates and oesophageal cancer was raised in a report of 54 adverse reaction case reports from the United States, Europe, and Japan received by the US Food and Drug Administration.1 Based on these cases, the authors of this study recognised the need for further epidemiological studies to confirm or refute the association between bisphosphonate use and oesophageal cancer.

What did the authors do?

To investigate whether users of bisphosphonates are at increased risk of oesophageal cancer but not stomach or colorectal cancer, Green and colleagues observed bisphosphonate use in patients with oesophageal, stomach, or colorectal cancer compared with controls in a nested case-control study. 2

A case-control study is an epidemiological study design that focuses on the aetiology, or cause, of disease rather than the treatment. In a nested case-control study, cases of a disease (in this instance, oesophageal cancer) are identified within an existing cohort and compared with controls (participants without the disease in question). This allows the two groups to be retrospectively compared in an attempt to identify potential risk factors or reasons why the disease might have developed. Case-control studies can only suggest an association between a risk and a disease; they cannot prove causality. This is because in addition to the exposure in question there are multiple other factors that might have been accountable. These are known as confounders.

Case-control studies sit further down the hierarchy of evidence than randomised controlled trials, systematic reviews, and cohort studies, but they are very useful for examining rare diseases. It is costly and time consuming prospectively to investigate the aetiology of rare diseases because it requires a huge number of participants and a long follow-up period. There is also no guarantee that enough people will develop the disease in question to allow reliable conclusions to be drawn. Case-control studies offer a quicker approach to investigating rare diseases by allowing the identification of cases and controls at the outset of the study so that they can be retrospectively compared. Medical records and interviews are some of the ways in which information is collected in case-control studies, and both of these methods are subject to potential recall bias. Although case-control studies might not be as robust as good quality primary studies, because they are cheaper, quicker, and less labour intensive, they offer an attractive way of investigating rare conditions. The study by Green and colleagues is a “nested” case-control study.2 This means the cases have been identified from within an existing cohort. The benefit of using an existing cohort is that investigators can take advantage of prospectively collected data rather than having to search for this from scratch. The cohort used in this study was the General Practice Research Database. This database contains information on 6 million people in the United Kingdom who are registered with an NHS general practitioner. The database contains prescription information, diagnoses, test results, hospital referrals, hospital admissions, and information on deaths as well as basic demographic and lifestyle data. The validity of case-control study results depends largely on the correct identification of cases, the quality of the data used, and the length of follow-up. One wrongly diagnosed case can be enough to have a large impact on the results, especially when the study is examining a rare condition where numbers of cases are low and follow-up is short.

For the purpose of Green and colleagues’ study, cases were defined as men and women aged 40 years and over with a diagnosis of oesophageal, gastric, or colorectal cancer recorded between 1995 and 2005.

The authors selected five control patients with no history of gastrointestinal cancer per case. In an observational study, it is important that investigators try to “match” controls to cases. Through carefully selecting control participants who are similar to cases at the outset, investigators can reduce the chance that any observations might be due to variables other than the exposure in question. In a case-control study, it is important that the exposure in question is clearly defined. Green and colleagues were concerned with exposure to bisphosphonates, which they defined as a record of at least one prescription of an oral bisphosphonate licensed in the UK for use in osteoporosis. The investigators calculated the relative risks and 95% confidence intervals for developing oesophageal, stomach, or colorectal cancer in relation to use of bisphosphonates. To ensure that any observed differences were due to the use of bisphosphonates, the investigators analysed the data taking into account pre-existing risk factors for oesophageal cancer, including smoking status, alcohol intake, and body mass index. This process is known as adjusting for confounders. They also looked at whether pre-existing gastrointestinal disease, osteoporosis, and use of non-steroidal anti-inflammatories, corticosteroids, or acid suppressants had an impact on the results.

To further improve the reliability of the results, the authors performed what are known as sensitivity analyses. This is where data are manipulated in various ways to see whether the results alter. The investigators in this study manipulated the data in three ways; by defining bisphosphonate exposure as two or more prescriptions; by restricting analyses to patients with complete data on alcohol, smoking, and BMI; and finally, by restricting data on bisphosphonate prescription, smoking, and alcohol use to that recorded more than one year before the index date. If results remain the same in spite of data manipulation, it is more convincing that the variable in question (in this case bisphosphonates) is likely to be responsible for the observed differences, thus strengthening the data. However, if there are differences in the results before and after sensitivity analysis, the results should be interpreted with caution.

In addition to looking at whether there was an increased risk of cancer with blanket use of bisphosphonates, the authors had detailed enough data and long enough follow-up to break this down further to examine whether the risk was associated with the number of prescriptions or the duration of use.

What did the study find?

Patients had a mean age at diagnosis of 72 years, and 43% were female. Three per cent of the study population had at least one bisphosphonate prescription during the follow-up period. To look at the degree of risk associated with bisphosphonate usage and the different types of cancer, the investigators calculated the relative risk. This is a measure of the risk of developing a disease in relation to a specific exposure. The relative risk value is a probability of the disease occurring in those subjected to the exposure in question compared with those without exposure. A relative risk value of 1 means there is no difference in risk between the exposed and non-exposed group. A relative risk value of greater than 1 indicates an increased risk in the exposed group compared with the non-exposed group, and a value of less than 1 implies that the exposure is protective. Relative risk values should be accompanied by a 95% confidence interval, which gives the two values between which you can be 95% certain that the actual value lies. For a result to be considered significant, the confidence limits should not cross 1. Green and colleagues found the adjusted relative risk of oesophageal cancer in those with one or more prescriptions of bisphosphonate versus those with no prescriptions was 1.30, 95% confidence interval 1.02-1.66; P=0.02. This means that any one patient who has been prescribed bisphosphonates has a 30% increased risk of developing oesophageal cancer compared with people who have never been prescribed a bisphosphonate. When broken down further it was found that this risk was significant only in those who had been prescribed more than 10 prescriptions or those who had prescriptions over a time period spanning more than three years. The excess risk was not altered by age, sex, smoking, alcohol intake, or BMI. There was no difference in risk observed between the various bisphosphonate drugs prescribed and no increased risk of stomach or colorectal cancer in participants who had used bisphosphonates. The relative risks for all cancers and subgroups are set out in the table⇓ below.

The sensitivity analyses yielded similar results, which strengthens the data.

What are the strengths and limitations of the study?

The investigators did what they could to minimise the likelihood of confounding by adjusting for several known risk factors for oesophageal cancer. However, it is impossible to account for all potential confounders, and it is important to remember this when interpreting results from observational studies. The study benefited from using a large, established database, which allowed the investigators to take advantage of comprehensive prospectively collected data on prescriptions and demographics with decent follow-up. Even though this dataset is likely to contain detailed and accurate information on the number of prescriptions issued, be aware that this does not mean all patients were compliant with the drugs prescribed. When prescription data are used, results must be interpreted cautiously as this does not necessarily correlate with the true compliance rate within the study population. Another issue for consideration is that the authors say that histological diagnoses were only available in 20% of cases. Although the authors didn’t find any difference in bisphosphonate associated risk between the different histological types of oesophageal malignancy, they acknowledge that this analysis was based on small numbers and might have been different if the histological data were more complete.

What does the study mean?

This study shows that patients taking oral bisphosphonates are at an increased risk of oesophageal cancer but not stomach or colorectal cancer. The increased risk is confined to those who have been issued with more than 10 prescriptions or who have had prescriptions for more than three years. Based on European and North American oesophageal cancer incidence in the 60-79 age group, the risk increases from 1 per 1000 to 2 per 1000 with five years use of bisphosphonate use, so the absolute risk overall remains low.

What are the implications of this study to medical students and clinical practice?

Case-control studies are useful for examining the aetiology of disease, but it can be difficult to translate the findings into clinical practice. To overcome this problem, editorials are often commissioned to accompany new research in an attempt to put the study conclusions into a clinical context. At the time this paper was published, another study investigating a similar clinical question using the same database was published in JAMA.3 Despite using the same data, the conclusion from the JAMA trial was that there is no increased risk of oesophageal cancer in patients taking bisphosphonates. An editorial published in the BMJ to accompany Green and colleagues’ paper explains that the differing conclusions are likely to be due to the shorter follow-up in the negative JAMA study (4.5 versus 7.7 years) and the larger sample size and consideration of other potential risk factors in the positive BMJ study.4 The editorial highlights how Green and colleagues’ study supports the association of an increased risk of oesophageal cancer in bisphosphonate users. However, even if the association is true, the overall incidence in the population remains low. From the point of view of medical students and junior doctors, the editorialist recommends careful consideration of the risk-benefit profile, careful inquiry about gastrointestinal symptoms before prescription, clear advice about administration, and what symptoms patients should be alerted to.

Hopefully, further studies (preferably prospective studies) might confirm or refute this association, but until then students and junior doctors should continue to follow current guidance about the use of bisphosphonates for osteoporosis, while bearing these cautions in mind.