How much bone for the buck? The importance of compliance issues in economic evaluations of bisphosphonates

Medical care usually entails careful consideration of benefits, risks and costs. Bisphosphonate therapy for the prevention and treatment of osteoporosis has significant benefits at moderate risks that come at costs which are somewhat higher than the price of a box of calcium tablets [1–5]. Prescription rates from the USA show the increased acceptance of bisphosphonate therapy. A total of 97% of all osteoporosis patients will leave their physician’s practice with a prescription, 73% of which will be for a bisphosphonate [6]. Recent economic evaluations of bisphosphonate therapy have shown that these antiresorptive drugs are cost effective for certain patient populations [7–12].

Is this really always the case? How cost effective is a treatment that may not be taken but instead takes up space in the bathroom cabinet (where drugs should not be stored in the first place anyway)? Did we forget about compliance? It is known that compliance with bisphosphonate therapy is often suboptimal [13]. The results of a recent analysis by McCombs and colleagues show that the mean unadjusted duration of continuous therapy is 245 days for bisphosphonates [13,14]. Furthermore, we know that high compliance with bisphosphonate therapy reduces the risk of a fracture by 16% [14]. Low compliance will therefore reduce the effectiveness of the intervention and consequently its cost–effectiveness.

The vast majority of all economic evaluations of bisphosphonate therapy (or of interventions for the prevention or treatment of osteoporosis in general) are studies that make use of decision-analytic models. Based on the results of a recent review on economic evaluations for the prevention and treatment of osteoporosis published between 1980 and 2004, 41 out of 42 studies were model-based [15]. This is not surprising. Osteoporosis is a chronic disease, the event rate (fracture rate) is relatively small, and the beneficial effects of therapy will not be effective until months after the onset of therapy. Decision-analytic models such as the frequently used Markov models, provide the possibility of easily projecting the course of a chronic disease over time, including the respective number of events and resource use [16,17]. However, since the model structure depends on the research question (and research group) and is naturally not predefined, it is up to the analyst to build a model in such a manner that it reflects clinical reality or the real world. As compliance may substantially jeopardize effectiveness, this shoud clearly be taken into account when a decision analytic model is constructed [18].

How has compliance been addressed with hitherto in published models? The striking result is almost not at all, at least not explicitly. In a review of decision-analytic modeling studies of the cost–effectiveness of interventions for the prevention of osteoporosis published in the years 2001–2004 [unpublished data], only four out of 17 studies addressed compliance issues either in the base-case or sensitivity analysis. However, most of these studies addressed the problem of poor compliance in the discussion sections of the papers. Frequently, it was stated that poor compliance may influence the result of the analysis and would lower the effectiveness of the intervention. It was then further argued that such reduced effectiveness would to some extent be offset by lower treatment costs [7,11,12]. This sounds intuitively right, however, we should be interested in total healthcare costs and not just treatment costs. A recent study by Sokol and colleagues has shown that for four chronic conditions (diabetes, hypertension, hypercholesterolemia and congestive heart failure), compliance is negatively correlated not only with hospitalization risk, but also total healthcare costs [19]. Therefore, although medical costs may be lower due to poor compliance, overall healthcare costs might be larger. This is due to a larger hospitalization risk in patients that comply poorly with their therapy. However, McCombs and colleagues do not find such a strong relationship between patient compliance with osteoporosis prevention/treatment and fracture risk [13]. The additional drug costs of US$266 for 1 year, when the patient fully complies with therapy, is only partly offset by the reduced costs for physicians (-US$56), hospital outpatient services (-US$38) and laboratory use (-US$9), and other hospital costs of -US$155 (total reduction of nondrug costs in case of compliance: US$258). However, it should be noted that a higher fracture rate is also associated with productivity costs and a reduced quality of life. It is very likely that when the full societal economic consequences are considered, the impact of compliance on cost–effectiveness would be even larger.

However, even if we exclude total healthcare costs from our analysis, there is something odd about stating that lower effectiveness through lower compliance is partly offset by lower treatment costs. Why do we not just simply say that we may overestimate the intervention’s cost–effectiveness (i.e., report too low incremental cost–effectiveness ratios) if compliance is suboptimal? Most researchers are aware of the problem of poor compliance. So why do we not find more studies that properly deal with it? The answer to this can be found in a recent study on the cost–effectiveness of the bisphosphonate alendronate, which stated that ‘compliance is not an easy issue to handle in economic evaluations’ [11].

It is difficult to estimate the level of compliance with bisphosphonates over time, and it is even more difficult to estimate the effect of poor compliance on the amount of relative risk reduction associated with bisphosphonate therapy. In order to be able to discuss the influence of poor compliance, we first need to establish what we mean by compliance or being compliant. One possible way that is frequent in the medical literature is to consider patients as compliant if they have medication available during a certain duration of treatment time (e.g., prescriptions are obtained to cover 80% of treatment time) [14]. As we do not know whether the patients are actually using the drugs they obtained, using this definition will overestimate the level of compliance, yielding a conservative estimate of patients’ actual compliance. Some patients will have their tablets ready but then simply forget to take them or intentionally decide to stop taking their medication for whatever reason (e.g., side effects or costs of medication).

Ultimately, it comes down to the question ‘how much compliance is enough for full effectiveness?’ [20]. This is determined by the drug’s pharmacodynamics (i.e., the drug’s dose–response relationship). There are drugs that require either substantially more or less than 80% of prescribed doses taken for full effectiveness, the widely used minimum threshold for considering a patient as being compliant [20]. Although bisphosphonates have now been used for many years, their complex pharmacokinetic/pharmacodynamic relationship is not yet fully understood [21]. Having said this, we should have a closer look at the data from McCombs and colleagues again [13].

McCombs and colleagues show that 42.7% of all patients in the first treatment year take their medication for less than 90 days and only 31% of all patients in their inital year of therapy have a level of compliance above 80%. These figures are based on prescription data of 3720 Californian patients who were prescribed bisphosphonates. In light of these data, it seems to be quite a strong assumption that a cohort of, for example, 5000 patients would take their weekly (or even daily) dose of bisphosphonates for 5 years without any interruption or break of therapy.

Another problem arises because we do not really know how patients do not comply. Some patients will just switch to other therapies, others will refuse to take any medication and a third group of patients will have an intermittent therapy that is just a little bit more intermittent than originally intended by the treating physician (i.e., they will forget to take their medication every now and then). One way to take these latter patients’ lower compliance into account is to assume that a reduced level of compliance of for example 80% will result in only 80% of the drug’s full effectiveness (e.g., 80% compliance to a treatment with a relative (fracture) risk of 0.6 under a level of compliance of 100% will result in an (increased) relative risk of 0.68 (compliance adjusted relative risk = 1 – ([1–0.6]*0.8) = 0.68). Admittedly, assuming a linear relationship between relative risk reduction and compliance is fairly arbitrary, but the analyst may also want to explore alternative relationships in a sensitivity analysis.

Currently, we do not have data to provide a standard approach for how compliance should be technically dealt with in analyses, but neither do we believe that omitting compliance issues from the analysis is the way forward. The absence of data in itself is not a justification for simplification [22]. Patients who switch therapies accrue no further costs and likewise obtain no further benefits if we do not assume the presence of any positive effect during the ‘offset time’. This seems to be a reasonable assumption since 37% of bisphosphonate patients switch to a second medication within their initial treatment year, which is a short time period for any clinical effect to become significant and large enough to have an effect during the ‘offset time’. Patients who just store their medication at home probably do not represent the largest part of patients.

In conclusion, what we have tried to emphasize in this editorial is that compliance may have a substantial impact on how much bone we get for the buck, and we therefore recommend its formal implementation in economic evaluations of bisphosphonates.