Predicting Medication Compliance and Persistency By: Jay Bigelow, President Amanda Rhodes, M.P.H., C.H.E.S., Vice President Behavioral Solutions MicroMass Communications, Inc. Introduction A widely recognized issue in health care is the high percentage of patients that do not properly follow the medication regimens prescribed by their physicians. This problem, generally referred to as non-adherence, has two components. The first concerns the degree of compliance, which describes how well a patient adheres to a medication regimen, i.e. whether the patient takes correct dosage of the medication according to the prescribed schedule. The second component is the degree of persistency, which refers to the length of time during which a patient is compliant with a medication regimen. In general, the rates of medication non-adherence are found to be between 40-60%[1-4], and can exceed 80% in certain patient populations[5]. Medication non-adherence has many negative consequences that result from subtherapeutic dosing, such as uncontrolled disease, poor health outcomes, and reduced quality of life. There are, however, some less apparent but perhaps more insidious effects of not adhering to medication regimens. For example, studies have shown that it is important for patients who take cholesterol-lowering statins to stay on their medication as interruption of therapy leads to significant risks in the occurrence of heart attack[6,7]. In addition to adverse health impacts, the cost of medication non-adherence is estimated to be $100 billion per year in the United States alone, including $30 billion in direct medical cost due to hospital admissions and unnecessary nursing home placement, and $70 billion due to lower productivity and premature death[8]. In the mean time, pharmaceutical companies sustain billions of dollars annually in lost revenue as a result of poor patient compliance and persistency. Given the enormous health and economic impacts of this issue, it is not only beneficial to patients but also in the interest of healthcare providers, pharmacoeconomic and health outcomes experts and pharmaceutical companies to identify all of the factors involved in medication compliance and persistency. From the perspective of a relationship marketing agency like MicroMass Communications, understanding what precipitates non-compliance is essential for devising messaging strategies that produce desired health behavior and outcomes. The consensus among researchers is that patient compliance and persistency with any specific medication involves a range of factors. Furthermore, reasons for non-compliance can vary dramatically from person to person even when the patients being studied belong to the same age, gender and socio-economic groups, have the same medical condition and take the same medication. Despite the findings of a large body of research, a common belief in the health care industry holds that the degree of adherence is mainly accounted for by the combined influence of perceived efficacy, side effects and the financial cost of treatment, and that behavioral drivers (such as a patient s attitudes, beliefs and barriers to treatment) hold little sway in determining whether a patient will be compliant and/or persistent with a therapeutic regimen. This assumption about behavioral factors lack of relevance, however, is contradicted by our experience at MicroMass. During our 10 plus years as a provider of messaging strategies and programs based on behavioral science we have found that the rate of medication compliance and persistency can be significantly improved by addressing behavioral drivers through patient communication programs. The success of MicroMass programs notwithstanding, the behavioral drivers that we targeted in any particular patient group covered all those suggested by existing health education theories. However, without the backing of a predictive model that would indicate which of those drivers would have the most profound effect in improving medication adherence, we had no way of knowing the relative importance of the individual drivers targeted, which ones had the greatest influence in terms of behavior change. Believing that a better understanding of behavioral drivers would lead to improved adherence and more streamlined messaging strategy, MicroMass conducted primary research to isolate the factors pertinent to compliance and persistency in specific therapeutic categories. This study directly tested the hypothesis that
behavioral drivers are important, and allowed us to rank the relative significance of all relevant drivers and to pinpoint those that can be addressed in a communication program. This type of model would facilitate the design of more effective communication programs tailored to individual patients. The study also provided the basis for constructing a quantitative model that predicts accurately, for any given consumer, the initial patient annuity value (PAV, i.e., the total value of a medication a patient purchases in a year) and the potential PAV if relevant drivers are successfully addressed. In addition to better helping patients who are already on medications, this research also can drive new patient acquisition. Insights gained from such study would enable pharmaceutical companies to identify among potential users of a medication those who have the highest value and to devise messaging strategies that preferentially target these individuals. Finally, the methodology exemplified in this study also can be used to assess the merit of an existing brand segmentation model. For example, the marketing team of a brand typically segments current and potential customers according to gender, age and other demographic criteria for the purpose of better directing promotional efforts. This sort of segmentation model is based on assumptions that may or may not be accurate. Findings obtained using the validated approach presented here, however, would indicate whether such models are useful, and would do so with a high degree of confidence since the judgment would be based on concrete data rather than on unproven assumptions. Selection of appropriate behavioral models Our study involved multiple brands in a single therapeutic category. The first step was the selection of appropriate behavioral models. Behavioral models are scientifically proven theories that describe the process of human behavior change in particular situations and indicate the factors that may affect this process. For example, the health belief model[10-12] is a widely used framework for understanding why a person takes (or does not take) certain preventive measures with regard to that person s own health, such as undergoing mammography screenings for breast cancer. The theory contends that the drivers influencing such actions include, among others, perceived susceptibility (i.e., belief as to how likely it is she will develop breast cancer), perceived severity (opinion of how serious the disease is), self-efficacy (confidence level in her ability to take action to prevent the disease) and cues to action (e.g., information that increases breast cancer awareness). A large number of health behavior models have been proposed in past decades. Each was formulated to address specific issues in health education and other programs that make use of behavioral interventions. Thus not all theories are applicable in any single situation, and no one theory is applicable to all situations. To determine which behavioral models are relevant to the degree of compliance and persistency with the medication regimens examined we: conducted an analysis of secondary research literature identified which models had been used in health education programs directed at patients suffering from conditions for which the medications under study are typically prescribed reviewed all relevant brand market research to fully understand the positioning and target audience. By combining the information gleaned from the literature and that from prior brand research, we were able to make educated choices of appropriate behavioral models and theories to blend together. These models then served as the foundation for determining which behavioral drivers would be tested in the survey. It should be emphasized that there is no one-size-fits-all approach for the process described here: choosing which behavioral drivers to test is performed on a case-by-case basis and demands a high level of expertise from the behaviorists involved and a clear understanding of the brand team s goals and objectives. Constructing the survey instrument and conducting the survey Once the list of behavioral drivers to be tested had been established, the next step was to determine the appropriate survey questions for assessing each driver. Survey questions were designed for drivers for which no previously validated instrument was available. All questions were then combined into a single survey. In addition to the queries for evaluating the behavioral drivers, the final survey also included questions that assessed two of
the three drivers that have dominated the conventional thinking of the healthcare industry, i.e., the patients opinion of how effective the medications are, and how serious they perceive the side effects to be. (Information about the third conventional driver cost of treatment was obtained from transaction data.) Lastly, as a standard practice, the survey contained questions (age, gender and online behavior) that would help profile the test population. The survey was sent to 175,000 people who had opted-in to receive additional pharmaceutical communications sponsored by the client. From this mailing 58,000 completed the survey, representing a response rate of 33%. The high response rate was in part due to the use of small but suitable incentives. Matching survey responses to transaction data As a prerequisite for further statistical analyses, survey responses needed to be linked to each respondent s degree of compliance and persistency. An objective measure of compliance and persistency can be achieved by examining the history of a patient filling and refilling prescriptions. This transaction data can be obtained from PBM companies (or any company that resells such data). With the assistance of a third party we matched the survey responses, at the individual level, to transaction data provided by a PBM company. The matching process was HIPAA-compliant (Health Insurance Portability and Accountability Act). Thus, although each of the matched data sets belonged to a single patient, none of the sets could be traced to a specific patient. This process resulted in a total of almost 6,500 matched data sets, which were used in subsequent statistical analyses. Analyses of data sets First, we determined for each patient in the analysis the degree of compliance and persistency with the medication regimens. The 6-month period immediately prior to the survey was set to be the time window during which these parameters were measured. Persistency was defined by whether a patient continued to fill a certain medication without exceeding a 30-day gap in therapy. Compliance was defined as the ratio of the total number of days of therapy received divided by the total number of days in the 6-month period. Next, we evaluated to what extent each of the tested drivers was present in a given patient. The final score for each driver by individual was expressed as high, medium or low. A regression analysis was performed to assess the degree of correlation of high and medium driver scores to medication compliance and persistency. The analysis was carried out for the entire sample, and separately for users of each of the prescription brands considered in the study. The results showed that several behavioral drivers were significantly correlated to the degree of medication adherence, thus validating our hypothesis that behavioral drivers are important. In addition, the three conventional drivers perceived efficacy, side effects and cost (co-pay) also affected medication adherence under most circumstances. A further interesting finding was that significant drivers varied somewhat according to brand, although the brands tested belong to the same therapeutic category. After the discovery of the statistically significant drivers, we ranked the relative importance of those drivers. The impact any given driver has on medication adherence was measured as the ratio of the average degree of compliance and persistency of the group of patients that had high or medium scores on that driver divided by the average compliance and persistency of the group that had low scores. The relative importance of drivers was thus ranked according to these impact ratios. Therefore, the statistical analyses conducted in this step not only identified which drivers were relevant to medication adherence, but also ranked those drivers in terms of their relative impact. Constructing and implementing a scoring matrix The goal of this study was to translate the statistical findings into tools to help fine-tune communication program messages for individual patients. To this end, we first determined, using a proprietary process, which drivers that were found to affect medication adherence can be modified through messaging, and that a modification of the driver(s) would have a positive impact on adherence. Obviously, there is little we could do to change drivers such as perceived efficacy, side effects or cost (co-pay). However, most of the significant behavioral drivers identified in this study are actionable.
After having identified actionable drivers, we next constructed a model that was referred to as a scoring matrix. The model serves two very useful purposes. First, it predicts, based on a patient s scores on all relevant drivers (actionable or non-actionable), the patient s degree of adherence and, therefore, the present value the patient represents to a brand; second, it predicts the patient s potential value to the brand assuming a modification of any or all of the actionable drivers would be achieved. To predict the value of a patient impact ratios associated with significant drivers and two other parameters are used. One is the value per patient averaged across the patient population taking the medication in question, which we refer to as the baseline PAV. Thus, if patients purchase, on average, $300 worth of this medication per person per year, then the baseline PAV is $300. The second parameter is an adjustment factor that indicates the ratio of the average compliance and persistency of the patients who have low scores on all relevant drivers divided by the average compliance and persistency across the entire patient population. We are then able to compute the present value of any specific patient by multiplying the following numbers: the baseline PAV, the adjustment factor, and the impact ratios associated with the patient s scores on all the relevant drivers. This type of calculation can be performed for patients with any combination of driver scores. The model indicates not only which behavioral drivers to target in messaging to a particular person, but also how much difference such intervention would make in terms of added value to a brand. The latter enables pharmaceutical companies to make informed decisions about how much marketing budget to spend on each audience segment. Furthermore, the ability of the model to predict a patient s value according to underlying behavioral drivers allows the targeting of high-value individuals by a brand s patient acquisition efforts. Implications This study represents a novel method of integrating three different areas of expertise: behavioral science, primary market research and predictive modeling. To the best of our knowledge, no work of this kind has been previously carried out in the field of pharmaceutical marketing. The results of this study validated our hypothesis that behavioral drivers have significant influence on medication compliance and persistency. It also disproved the long-held assumption that perceived efficacy, side effects and cost of treatment are the only factors significantly affecting compliance and persistency. Since the study was conducted on a much larger scale than most previous market research in similar areas, the conclusions drawn from it also carry greater statistical significance. The greatest assets of this study are the abilities to identify specific drivers underlying adherence, to rank those drivers in terms of the irrelative impact, and to determine which drivers should be targeted to achieve greater compliance and persistency. The results of such study would then allow the construction of models that are able to predict a patient s degree of adherence simply from the patient s scores on a few drivers. This type of model can be used to determine the present and potential values of any individual or any group of patients; it can also guide the design of employable programs that are fine-tuned to the level of individual patients. The process presented here has great flexibility and can be adapted in many different ways. For example, more than one brand of medication can be studied at a time (as we have done here), if the products belong to the same therapeutic category. Studying multiple competitive brands simultaneously allows a pharmaceutical company to better understand the market of its own brand as well as that of its competitors. This would facilitate the design of switch messaging. Furthermore the method can be used to test the validity of existing segmentation models for a brand. We found that the intuitive assumptions on which such segmentations are based were, in some cases, invalidated by more rigorous study. In addition to its application in improving compliance and persistency, an area usually associated with patient retention, the methodology described here will have significant impact on new patient acquisition. It allows precise identification of the patients that have the highest potential value to a brand, and identifies the factors that drive these individuals to action. This sort of information is crucial for designing more precisely targeted and effective messaging strategies to acquire new customers.
It should be emphasized that, because behavioral drivers underlying adherence generally vary a great deal across different therapeutic categories, the results from research in one category cannot, and should not, be applied to another category. The procedures entailed in the present study thus need to be repeated for any new category. Given its scale and complexity, this sort of research costs time and money. However, the effort will, in most cases, be well rewarded in terms of its benefits across all brand marketing efforts from patient acquisition through retention.