Understanding Patient Adherence and its EconomicConsequences

The real world effectiveness of treatments depends in part on the adherence of patients to the regimens prescribed by their physicians. Considerable effort has gone into research assessing adherence and how to improve it in the following ways:

• Adherence to regimens for chronic diseases, especially those such as hypertension and diabetes for which patients are initially asymptomatic, has been documented as severely suboptimal. Research has demonstrated that this in turn has contributed to worse long-term clinical outcomes and associated quality of life and economic burden.
• Adherence has also been shown to be sub-optimal in symptomatic diseases, attributable in part to a combination of patient clinical and demographic characteristics as well as health care system and provider characteristics. Studies to assess these barriers to and facilitators of positive adherence have grown more numerous recently, and have led to policies and programs to lower those barriers and facilitate better adherence.
• As well, improvements in drug delivery have made possible formulations that make adherence simpler: e.g., less frequent dosing, oral administration rather than injection. These newer technologies are also being evaluated to confirm their hypothesized advantages.

As these issues have been recognized, the number of studies to understand adherence to prescribed treatments, the factors associated with greater or lesser adherence, and the consequent clinical, quality of life, and economic outcomes have increased greatly in the last few years. A MEDLINE search revealed that the number of publications addressing adherence rose by more than 60% from 1,610 in 2000 to 2,680 in 2004. Some of these publications reported the results of prospective studies, typically either randomized clinical trials or patient registries, but also case series or single- center experience reviews. Other publications reported the findings of retrospective studies, frequently of claims based database studies, or sometimes of models presenting parallel assessments of clinical trial results synthesized using a model framework.

Across these many studies, one can identify numerous challenges in the measurement of and interpretation of adherence. At the 8th European ISPOR congress held in Florence, Italy in November 2005, I presented a workshop on some of these challenges to effectively designing prospective or retrospective studies addressing adherence, and the responses to overcome these challenges. This short article reviews aspects of the workshop’s presentation and discussion. In particular, this paper focuses on several methodological and analytical issues commonly found in research on adherence and its economic consequences. Consideration of each of these points by researchers studying adherence should improve the design of their analyses and lead to more informed cost analyses as well.

First, appropriate measures of adherence need to be carefully selected and this heavily depends on the disease or therapeutic area being studied. Many studies in numerous therapeutic areas have evaluated adherence using a traditional measure known as “medication possession ratio” (MPR), or more recently “time to treatment change” employing survival analysis techniques. For chronic diseases such as hypertension and diabetes, a plausible scenario is that despite an immediate increase in medication acquisition costs due to higher levels of medication adherence, overall healthcare costs with respect to hospitalizations, physician visits, and other drug costs may decrease in the long run (e.g., 2-3 years) as adequate levels of adherence allows patients to achieve desired level of clinical endpoints. Therefore, treatment discontinuation is suggestive of poor adherence. In contrast, for other diseases in which patients can be cured or relieved of symptoms, treatment discontinuation as a measure of adherence may not necessarily constitute non-adherence to treatment; rather, it may reflect relief of symptoms or achievement of the required clinical goals. As a result, patients who benefit from symptom relief sooner will have a shorter treatment period. It would be nice if treatment periods were equal across all patients, and positively associated with level of adherence. Unfortunately, that may not be typically true. Those with low adherence may have prolonged disease and eventually higher overall healthcare costs. Alternatively, those with low adherence may simply not have access to health care, and their use of health care and associated costs will be lower. Studies of differently insured populations, or different age, race and ethnic groups have found this to be the case. Overall, then it is important to understand both overall trends and levels of adherence, and also to disaggregate cohorts to better understand the adherence levels, disease experience and associated costs of these sub-cohorts.

Secondly, to achieve more complete understanding of adherence and associated health care use and costs, the main cohorts should be studied not only overall, but also disaggregated. I have found it useful to categorize patients into four subcohorts based on their use of or adherence to a drug or other intervention. In mathematical terms, the cohort could be defined this way:

Total cohort T = T1 + T2 + T3 + T4

Where
T1 = subjects fully adherent with therapy (patients not stopping or switching medications)
T2 = subjects sub-optimally adherent with therapy (patients not stopping or switching medications but with a MPR below a certain level)
T3 = subjects that switch therapies
T4 = subjects that discontinue or stop therapy

Consequently, costs may be similarly disaggregated, as in the following equation:

Total health care cost PT = P1 * T1 + P2 *T2 + P3 * T3 + P4 * T4

Where,
P1= average total cost for T1
P2= average total cost for T2
P3= average total cost for T3
P4= average total cost for T4

An assessment of overall health care costs associated with use of different drugs having different adherence profiles is conceptually easier to understand with these equations. One can more readily understand both the different levels of adherence of multiple cohorts and the costs associated with differential adherence. In evaluating the adherence of different cohorts following two (or more) alternative therapeutic regimens, one can identify the different proportions of the cohorts with varying levels of adherence and the costs of the associated level to see how they together affect overall costs. One can more readily identify the pharmacy impact as well as the overall health care utilization impact. This type of analysis provides a complementary and more complete look at adherence than that which is provided by a summary score such as MPR or time to discontinuation. Analyzing both types of outcomes provides physicians and policy makers a more complete look at the options they face and provides some insight into possible ways of improving adherence and optimizing cost. the median duration for Cohort B is shorter than that for Cohort A. However, the wide variation of therapy duration in Cohort B creates uncertainty about the overall value of this therapy.

Thirdly, the impact of medication adherence on health care costs may have a lag time, so that adherence levels are not necessarily contemporaneous with either improvement in clinical outcomes or costs. This may be the case for multiple reasons, any one or more of which may be the case in the disease that you are evaluating. For example, in some therapeutic areas (e.g., depression), this may reflect the slow action of the medication that does not show effect for two to four weeks or more. In other therapeutic areas, it may be more reflective of the chronic nature of the disease. As an example, adherence in the short term may have an impact only on long-term clinical or economic burden, but not on short-term clinical or cost improvement. A third scenario is that a disease may manifest itself differently and/or the drug may improve outcomes in a sufficiently varied manner temporally that researchers need to disaggregate the drug-specific cohorts by clinical and/or demographic characteristics to better identify what is happening in which patient cohorts when. Multiple sclerosis is one such disease that manifests itself differently, as do the alternative treatment regimens. Their effects are seen differentially over time. On a more basic level, this suggests an econometric analysis where, for example, total cost (or any cost component such as hospitalization or ER visits) may be a function of adherence in a prior time period. Accordingly, an appropriate lag time should be explored.

Fourthly, and in a similar fashion, relatively little attention is typically paid to distributional effects of treatment duration measured by “time to discontinuation.” For analysis of disease that can be “cured” or relieved of symptoms in a relatively short term, treatment discontinuation may reflect symptom relief or attainment of desired clinical endpoints, so that short treatment duration could be construed as positive. In this case, considering the variation of treatment duration is critical in an adherence study. Although mean or median duration of one therapy may be shorter than another, it could potentially indicate a quicker recovery, and a different distribution of therapeutic effectiveness. For example, as shown in following histograms of the duration of therapy for Cohort A and Cohort B,

Fifthly, the new availability, or launch, of a new product will affect use of, and potentially adherence with, existing products. Concurrently, those patients who are immediate adopters of the new drug may not be representative of the general cohort of patients most appropriate for using the new drug. Consequently, for both of these reasons, the immediate period around the time of a product launch is not the ideal or even appropriate time to assess adherence, either of the existing products or of the new product. In the first instance, for example, let us assume that we compare two cohorts, one treated with a pre-existing drug (Drug A) and the other treated with the new drug (Drug B). When assessing the time to switch, one may find that many patients treated with the pre-existing Drug A switch more readily to new Drug B because for whatever reason they are not fully satisfied with Drug A and want the newly available and/or novel Drug B. Characteristics of the disease and the drugs (e.g., are they suboptimal?) may make this more or less a problem. In the second instance, adopters of the new drug will tend to include a disproportionate percentage of those who are refractory to the existing therapy. One would then suppose that their experience on the new drug and that includes adherence, could be either worse or better, than other patients. With this in mind, an adherence study will be more appropriately timed other than during the perilaunch period.

Overall, medication adherence will remain an important topic for physicians and other health care providers, for researchers and of course for patients. The ability of researchers to more effectively analyze their data to permit more informed interpretations of what truly is going on will lead to better designed studies that should more effectively assist physicians and policy makers, and ultimately improve patient adherence and outcomes.

  Issues Index | 2006 Issues Index