Patient-Accessible Tool for Shared Decision Making in Cardiovascular Primary PreventionCLINICAL PERSPECTIVE
Balancing Longevity Benefits Against Medication Disutility
Background—Primary prevention guidelines focus on risk, often assuming negligible aversion to medication, yet most patients discontinue primary prevention statins within 3 years. We quantify real-world distribution of medication disutility and separately calculate the average utilities for a range of risk strata.
Method and Results—We randomly sampled 360 members of the general public in London. Medication aversion was quantified as the gain in lifespan required by each individual to offset the inconvenience (disutility) of taking an idealized daily preventative tablet. In parallel, we constructed tables of expected gain in lifespan (utility) from initiating statin therapy for each age group, sex, and cardiovascular risk profile in the population. This allowed comparison of the widths of the distributions of medication disutility and of group-average expectation of longevity gain. Observed medication disutility ranged from 1 day to >10 years of life being required by subjects (median, 6 months; interquartile range, 1–36 months) to make daily preventative therapy worthwhile. Average expected longevity benefit from statins at ages ≥50 years ranges from 3.6 months (low-risk women) to 24.3 months (high-risk men).
Conclusion—We can no longer assume that medication disutility is almost zero. Over one-quarter of subjects had disutility exceeding the group-average longevity gain from statins expected even for the highest-risk (ie, highest-gain) group. Future primary prevention studies might explore medication disutility in larger populations. Patients may differ more in disutility than in prospectively definable utility (which provides only group-average estimates). Consultations could be enriched by assessing disutility and exploring its reasons.
The initiation of lifelong primary prevention therapy for cardiovascular disease in a high-risk patient should be based on a shared decision-making process between patient and doctor following the clear presentation of appropriate information, including the quantification of the risks and benefits expected from treatment and the cost and inconvenience (disutility) to the patient. This ideal scenario is almost never achieved.
Editorial see p 2500
Clinical Perspective on p 2546
Currently, primary prevention practice focuses on risk stratification by using population-based statistical estimates to determine which individuals would have most to gain from preventative therapy.1 Doctors are documented to view risks differently from patients, and both have difficulty in evaluating, perceiving, and conveying risks and benefits in an easily understood manner.2–5 The benefits of primary prevention are thus often presented to the patient without formal quantification of the cost, harms, or inconvenience they might incur. However, patients do understand risks and trade-offs3 and trust doctors more when presented with numeric information than when given vague interpretations of risk.6
Previous interventions, aimed at improving adherence, have used new methods to convey cardiovascular risk rather than tackling the underlying reasons why people stop medication. The focus has been on individual counseling, and on quantitative and graphical displays, or the use of imaging techniques such as coronary CT scans to improve risk perception.7–9 These are based on the principle that better risk perception will lead to higher adherence and persistence with primary prevention therapy.10–12
Patient inconvenience, or medication disutility, has rarely been taken into consideration when initiating therapy. Knowing one’s risk to be high does not necessarily mean that one will, must, or even should take a preventive step. Taking action depends on many factors, and a large part of a patient’s resistance to treatment involves the reluctance to embark on a lifetime of medication. Statins are cost-effective for most persons with coronary heart disease risk factors if they do not mind taking a pill daily.13–17
When medication disutility is incorporated into the risk-benefit equation, it becomes clear that the cost-effectiveness of statins is extremely sensitive to medication disutility. However, despite its crucial importance in determining the incremental cost-effectiveness ratio, medication disutility data are scarce.14 Because of the lack of data, guideline writers have had to work on the basis that medication disutility is negligible. Cost-effectiveness analyses have typically used base case estimates of zero disutility and covered up to 0.01 or 0.02 in sensitivity analyses.13,14,17 Expressed as an absolute lifespan gain, for current English life expectancy at age 50 years, this translates to covering in sensitivity analyses the possibility that patients may be willing to give up a lifespan as large as 3.6 (or at most 7.2 months) to avoid medication. The analyses highlight that conclusions are exquisitely sensitive to this value, but the data on which to base an estimate are limited.
We do not know how close to zero medication disutility is. Nor do we know whether its distribution is fairly narrow, in which case a single value may be suitable for use in disease prevention decisions for all, or whether the distribution is wide, in which case it may be advisable to assess disutility within individuals in clinical practice.
Our study is the first to attempt to quantify the spectrum of individual medication disutility for primary prevention in a sample of the general population. We juxtapose it against the spectrum of expected longevity gain from the initiation of statin therapy across the same general population.
Medication disutility was assessed in a random sample of the general population of London by face-to-face interviews with the use of a structured questionnaire. Medication disutility has been assumed to lie between 0 and 0.001 in the time trade-off studies used in previous economic calculations,15–17 which roughly translates to being willing to give up between 0 and 5 months of life to avoid taking a daily medication. We designed our study to be able to estimate the proportion of subjects having medication disutility of >6 months, with 95% precision and a confidence interval of ±2%, even if the actual proportion of subjects in the population with this level of medication disutility was as small as 5%. Power calculations were based on the assumption that medication disutility would be normally distributed in the population. The sample size required on this basis was a minimum of 300 participants. We planned to recruit 360.
Survey participants were approached in public thoroughfares in London, on the basis that they would potentially be the target population for cardiovascular screening and primary prevention. Participants were approached and recruited on 3 days in October and November 2010. Members of the public were approached until 360 agreed to participate.
To focus the survey on medication disutility and minimize other potential sources of low compliance such as cost, subjects were asked to imagine an idealized tablet that was available at negligible cost, with no need for prescription, nor medical supervision, nor follow-up blood tests. They were also asked to assume that the tablet would have no side effects and could be started or stopped at will with no consequence other than receiving only partial benefit.
Disutility was assessed by initially asking subjects whether gaining an additional day of expected life would have sufficient benefit for them to commence lifelong therapy with the idealized tablet. If the answer was negative, then the subjects were asked if an additional 10 years of expected life would suffice. If the answer was positive, medication disutility was assumed to lie in the interval between 1 day and 10 years. This range was progressively narrowed by using a binary tree (maximum 6 further steps) to reach the benefit required by each subject to offset their personal medication disutility.
The algorithm was constructed to approximately halve the time interval at each step, thus aligning the time points approximately evenly on a log scale. The speed of completion of the algorithm was confirmed by pilot testing and, on average, took <1 minute. Subjects who indicated that 10 years of longevity benefit would be insufficient were classed as having an extreme medication disutility. Demographic information on age, sex, employment status, current use of medication, and previous heart attack or stroke were also sought. The full questionnaire is shown Appendix I in the online-only Data Supplement.
Survey data were summarized by using simple measures of central tendency (mean and median) and spread across quartiles for each age and sex group. The distribution of medication disutility was also examined visually to assess whether it followed a normal distribution and whether it had the same shape in each age and sex group. Differences on tablet disutility across sex and age were tested by using parametric and nonparametric tests for both.
The survey was indicated by the local Ethical Committee chair to not require Ethical Committee Approval, because it assessed attitudes to an imaginary medication and was performed on members of the general public without collection of personally identifiable information.
Paddington Life Expectancy Gain Charts
We calculated the expected average increase in life expectancy due to the initiation of statin therapy for men and women with different levels of baseline risk with the use of standard multiple-decrement life table methods.18 Baseline life expectancy was based on all-cause and cardiovascular mortality rates for England and Wales in 200519,20 obtained from the Office of National Statistics UK. These rates were then decremented for high-risk groups according to the risk level induced by different permutations and combinations of the following risk factors: tobacco exposure, systolic blood pressure, total cholesterol, age, and sex. The size of the decrement for each age-sex-risk combination was calculated by entering values into the Systematic Coronary Risk Evaluation (SCORE) algorithm21 recommended by the European Heart Association for risk stratification and obtaining the percentage increase in mortality for each group. The SCORE algorithm compares each risk factor combination with the national average. Data on the national average mean and the distribution of blood pressure, smoking status, and cholesterol were obtained from the QRESEARCH database (2005) that includes data on >13 million patients spread throughout the United Kingdom.22
Data on diabetes mellitus have not been collected uniformly in SCORE study cohorts. Thus, people with diabetes mellitus were included in the general SCORE database used for the development of risk functions. However, because of nonuniformity in the ascertainment of diabetes mellitus, diabetes mellitus was not included as a dichotomous variable into the SCORE risk function.21 We have followed the same method for the decrementation of life expectancy in diabetic subjects in this study.
Blood pressure, total cholesterol, and smoking status above the national average level were considered to act multiplicatively to increase cardiovascular risk as per the SCORE algorithm. All in all, 40 different age-sex-risk combination tables were calculated to obtain values of expected longevity benefit for a full spectrum of risk groups (see Appendix II and Table I in the online-only Data Supplement for details). The design of the Paddington tables was kept as similar as possible to the SCORE charts, displaying, instead of a 10-year risk of fatal cardiovascular disease, the average longevity benefit (in months) that a patient can expect to gain by starting lifelong therapy with statins.
The percentage reduction in cardiovascular mortality with statin therapy was obtained from a meta-analysis of trials of lipid-lowering agents in primary prevention populations.23 For each cardiovascular risk group, life tables were then recalculated with the statin effect. The difference between baseline life expectancy and life expectancy with the statin therapy was taken as the average expected longevity benefit. The youngest age at which the initiation of statin therapy was modeled was 50 years. The spectrum of cardiovascular risk modeled was based on the distributions of blood pressure, cholesterol, and smoking in the UK population; thus, the spectrum of longevity benefit represents the average distribution of life-years gained from statin therapy in the UK population.
The Table shows the baseline characteristics of survey respondents. Three hundred sixty participants were recruited after 379 members of the public were approached. The distribution of medication disutility expressed as longevity gain desired by an individual to offset the inconvenience of taking a lifelong preventative tablet is shown in Figures 1 and 2 and Appendix III in the online-only Data Supplement. Two-thirds of subjects had medication disutility >1 month, and 12% had extreme medication disutility (desiring ≥10 years predicted increase in life expectancy to adhere to therapy). Near-zero medication disutility (<1 month longevity benefit required) was expressed by 34% of subjects. There was no relationship between sex and disutility (31±42 months in males versus 26±38 months in females, P=0.30 by t test, P=0.40 by Mann-Whitney U test). There was no relationship between age and disutility: Pearson correlation with age was 0.04 (P=0.42); with square root of the age, it was −0.01 (P=0.79); and Spearman rank correlation with age was −0.01 (P=0.79; Figures 1 and 2).
Tables of expected lifespan gain according to age, sex, smoking status, blood pressure, and cholesterol level of the subject are shown in Figure 3. The shading on the chart corresponds to the increase in group-average life expectancy for a notional large group of patients with that specified cardiovascular risk profile starting lifelong statin therapy. These life expectancy gains are meaningful only for the group as a whole, as is the case for risk percentages that are also sometimes displayed in this way. In practice, a small proportion of patients will gain the lion’s share of the extra lifespan, whereas a large proportion will gain no extra lifespan, as shown in Appendix IV in the online-only Data Supplement. From the age, sex, smoking status, blood pressure, and cholesterol, it is not possible to be more specific as to whether a particular patient will gain. Even if a trial were conducted, each individual patient could only be in 1 arm, and it would not be possible to pinpoint whether an individual patient had personally gained or not. The value represents only the mean for patients with that particular risk factor profile.
Figure 4 shows the frequency distribution of medication disutility (Figure 4, Top), juxtaposed against longevity benefit from statin therapy (Figure 4, Bottom). The calculated longevity benefit with statin therapy ranges from 5.5 months to 24.3 months in males, and from 3.6 to 18.2 months in females depending on the individual cardiovascular risk profile. Ninety-nine percent of the UK population will gain <24.3 months of additional life as a result of lifelong primary prevention with a statin, whereas 1% has a risk profile that allows them to gain more than this. Individual-subject medication disutility has a wide distribution in our survey population, ranging from <1 day to >10 years.
Figure 5 shows the expected distribution of longevity benefit in the English population resulting from distribution of total serum cholesterol (Figure 5A), systolic blood pressure (Figure 5B), smoking in the general population with all other risk factors held constant (Figure 5C), and the distribution of total cardiovascular risk using all 3 variables combined (Figure 5D). For Figure 5A through 5C, the distribution of longevity benefit with statin therapy was calculated allowing a particular risk factor (cholesterol, blood pressure, or smoking status, respectively) to vary with a prespecified distribution (the distribution of that risk factor in the population in the United Kingdom), while the other risk factors were held constant at the population mean. The distribution of longevity benefit for total cardiovascular risk was calculated by using all 3 variables combined in an aggregate risk score with the use of the SCORE algorithm.21
The implicit assumption in guideline development and clinical protocols for primary prevention of cardiovascular disease, namely that medication disutility is zero or near zero, may not be sound. Much more work remains to be done to develop evidence-based approaches to account for medication aversion during clinical encounters. In our simple study, even for an idealized tablet, more than one-quarter of individuals have medication disutility that exceeds the group-mean lifespan gain from statin therapy calculated for a very high cardiovascular risk group.
A simple calculation of averaged expectation of benefit versus disutility might suggest that the addition of even such an idealized agent would not be perceived by that individual patient to present a net gain. Whether they would judge the situation differently, if it were made clear that some patients would gain a great deal of lifespan while many gained none, is unknown and might be an important question to explore in future studies.
Prevalence of Medication Disutility in the General Population
The prevalence and degree of significant medication disutility in the general population, which is the target population of primary prevention, may often be much greater than previously assumed. The medication disutility curves (Figures 1 and 2) are not normally distributed, but centrifugal, with a standard deviation 1.5 times the mean. Nearly half of the population has disutility greater than double the median or less than half the median. The shape of the medication disutility distribution curve seems similar across age groups, suggesting that its shape is genuine and that ageing with the associated perceived nearness of mortality did not have a large effect (Figures 1 and 2).
Medication disutility varies dramatically from person to person to a much greater extent than estimated cardiovascular risk between individuals. Clinical practice evaluates risk factors by using statistical estimates to determine whether taking a statin is worthwhile, but the interindividual variation in medication disutility, which appears to have a greater effect on net benefit for individuals, is rarely addressed. This variation between individuals in the size of medication disutility is greater than the effect of variation in any one of the common risk factors used to determine thresholds for treatment (Figure 5).
Even if primary prevention guidelines were revised to incorporate a nonzero value for medication disutility, there is no single value that could reasonably be entered because disutility varies to such an extent between individuals, much more so than utility. If our data are representative, then alongside assessing blood pressure, cholesterol, and smoking status, it may be informative to assess individual medication disutility and explore its reasons.
Faced with a patient with high expected lifespan gain from preventative therapy but even higher medication disutility, the clinician should not simply withhold therapy. Equally, however, clinicians should not simply prescribe and assume that the medication will be taken. High disutility could instead initiate the exploration of its underlying reasons.
Use of an Idealized Tablet to Assess Medication Disutility
We were keen to determine the lower limit of medication disutility and therefore used a hypothetical intervention to assess disutility rather than a real intervention that might have an adverse reputation. The hypothetical medication enhanced lifespan without having the 4 principal drawbacks of primary prevention medications: cost to the patient, inconvenience of obtaining a prescription, perceived loss of autonomy to stop and start at will, and adverse symptoms. The removal of these barriers improves compliance with medical therapy for chronic diseases.24 With real drugs, the possibility of side effects, the inconvenience of having to obtain prescriptions, and the nonzero cost mean that the distribution of disutility is likely to be greater than the values we obtained, and the spectrum of values might be wider.
Our study should therefore be considered only a lower limit on medication disutility. Nevertheless, it identifies that disutility is not near zero and is not trivial in comparison with the benefits offered by a medication such as a statin. To translate this concept into clinical practice, further studies with questionnaires specifically designed to investigate real medications used in primary prevention would be needed. Such a design, specific to the individual agent, and a particular cost and arrangement for prescription, will give a more complete picture of real-life medication disutility in a particular clinical context.
Study Limitations and Future Study Design
We did not collect individualized risk factor data on the subjects in our survey and therefore are unable to plot a joint utility-versus-disutility distribution at an individual level. This would only have been possible with detailed background information (including the measurement of blood pressure and measurement of blood lipids). We did not impose this step because we wanted this survey to be broadly representative of the general population and not only those willing to participate in a research study. Thus, it is important to note that the longevity benefit distributions in Figure 5 describe the general population and not the particular subjects in this survey. We cannot exclude the possibility that our sample of subjects might not be representative of the general population. Furthermore, comparing the individual medication disutility with expected life-year gain can be problematic as the difference becomes significant, especially at the individual level. When using this questionnaire in real life, a physician should make clear that a calculated increase of 1 year in life expectancy is an estimate that is based on an average of lifespan gain among subjects. To make this difficult concept easy to understand for every individual, the physician could offer a page with 3 examples of how, among a group of 10 people with an average increase of 1 year, individual gain may vary from the mean (Appendix IV in the online-only Data Supplement).
Our questionnaire was a very simple form of the time trade-off method. It was aimed to be brief to allow us to sample the general population and minimize the possibility of examining only an unrepresentative subset biased toward an interest in health. Our choice of survey design achieved a 95% participation rate. In ultimate clinical practice, with a patient voluntarily engaging in a consultation and therefore already showing some level of commitment to the questioner, a more comprehensive tool would be appropriate.
We assessed medication disutility without assessing the individualized expected lifespan. It is possible that people who are formally told that their remaining expected lifespan is short might have less medication disutility. However, in our data set, age – known to the public to be the most powerful determinant of mortality risk – did not affect medication disutility. Future studies using individualized utility calculations would be able to test this hypothesis.
It is likely that a participant’s personal disutility may be influenced by context and situation.25 For example, if we had questioned patients within a general medical practice or a hospital outpatient department, then their response may have been influenced by the many health-related cues nearby. We cannot assume that the disutility assessed in a public space is equivalent to the disutility that would be assessed in a primary prevention scenario. Future studies are needed to assess medication disutility in patients attending a primary care service for screening and being considered for preventative treatment.
Despite our request to imagine an ideal medication accessible without effort and causing no side effects, participants’ responses may nevertheless have been colored by an expectation of a high rate and magnitude of side effects, for example, through non–placebo-controlled reports in the mass media.
Our survey had an upper limit on medication disutility of >10 years, which prevents us from being able to subclassify subjects beyond this ceiling. However, from a practical point of view, knowing exact disutility numerically when it is already above the maximum achievable longevity benefit may not be so important as recognizing that subjects with such strong medication disutility do exist.
Because mortality rates change over time, the survival depicted in any period life table will not perfectly reflect the true survival experience of a cohort. For example, secular improvements in health mean that actual life expectancy of cohorts is often longer than that predicted by using period life tables constructed by applying present-day survival rates across age groups. Furthermore, life expectancy varies from country to country and cohort to cohort, so that Paddington tables might need to be reconstructed for different countries and cohorts.
Our sample is limited to North West London, which may not be representative of other areas in the United Kingdom. However, survey participants were drawn from the general population, which is the target population of primary prevention therapy. To minimize intrusiveness, we did not ask subjects their ethnicity, but we did approach subjects without regard to their apparent ethnicity. Census data show that the general population of London is more ethnically diverse than most of the rest of the United Kingdom, with 58% being white British, 11.3% other white, 13.3% South Asian, 10.6% black, 1.5% Chinese, and 5.5% mixed or other.26 The consistently large variation in medication disutility in both sexes across all age groups suggests that distribution is genuinely wide. Interviewing subjects in other cities is likely to make the distribution not narrower but wider.
Individual medication disutility may be fluid over time, for example, being influenced by a personal heart scare or a cardiovascular event in a friend or family member. Mass media reports may also be unhelpful because, without the benefit of placebo control comparison,27 the extent of genuine incremental side effects can easily be overestimated.
Finally, our data reflect medication disutility in a primary prevention cohort, and we did not assess the impact of cardiovascular events on medication disutility in secondary prevention. Only 1 individual in the survey had a previous cardiovascular event. Further studies are needed to investigate the longitudinal behavior of medication disutility to determine how often medication disutility should be reassessed.
The tables presented in this study are designed to allow both patient and doctor to compare the risk and benefit of preventative tablet therapy to determine an average expected net benefit for a notional group of similar patients with the use of a mutually understood metric of lifespan gain. High disutility in an individual might prompt an exploration of the underlying reasons, and enhancing the interaction between patient and clinician in this way might strengthen the consultation.
Guidelines specifying a risk threshold for treatment may have been derived from a tacit assumption of near-zero medication disutility, which may not be representative for many subjects. Future public health research could explore more advanced methodologies, because our simple medication disutility assessment takes only a minute, less than the time taken to measure cholesterol and blood pressure.
Although still at an early stage, individualized quantification and discussion of medication disutility, and parallel methods of describing group-average preventative benefits, might bring us closer to primary prevention that is truly personalized.
The online-only Data Supplement is available with this article at http://circ.ahajournals.org/lookup/suppl/doi:10.1161/CIRCULATIONAHA.113.007595/-/DC1.
- Received January 18, 2013.
- Accepted March 19, 2014.
- © 2014 American Heart Association, Inc.
- Robson J
- Nexøe J,
- Gyrd-Hansen D,
- Kragstrup J,
- Kristiansen IS,
- Nielsen JB
- Spiegelhalter D,
- Pearson M,
- Short I
- Gigerenzer G,
- Edwards A
- Gurmankin AD,
- Baron J,
- Armstrong K
- Asaria P,
- Francis DP
- Lipkus IM
- Wells S,
- Kerr A,
- Eadie S,
- Wiltshire C,
- Jackson R
- Lazar LD,
- Pletcher MJ,
- Coxson PG,
- Bibbins-Domingo K,
- Goldman L
- Greving JP,
- Visseren FL,
- de Wit GA,
- Algra A
- Preston SH,
- Heuveline P,
- Guillot M
- 19.↵Mortality statistics: Cause, England and Wales, series dh2, no. 32, 2005 [Internet]. Table 1 estimated resident population: single years of age and sex. http://www.ons.gov.uk/ons/publications/re-reference-tables.html?edition=tcm%3A77-39675. Accessed April 15, 2014.
- 20.↵Mortality statistics: Cause, England and Wales, series dh2, no. 32, 2005 [Internet]. Table 2.9 deaths: underlying cause, sex and age-group: Chapter IX Diseases of the circulatory system. Available from: http://www.ons.gov.uk/ons/publications/re-reference-tables.html?edition=tcm%3A77-39675. Accessed April 15, 2014.
- Conroy RM,
- Pyörälä K,
- Fitzgerald AP,
- Sans S,
- Menotti A,
- De Backer G,
- De Bacquer D,
- Ducimetière P,
- Jousilahti P,
- Keil U,
- Njølstad I,
- Oganov RG,
- Thomsen T,
- Tunstall-Pedoe H,
- Tverdal A,
- Wedel H,
- Whincup P,
- Wilhelmsen L,
- Graham IM
- Hippisley-Cox J,
- Coupland C,
- Vinogradova Y,
- Robson J,
- May M,
- Brindle P
- Taylor F,
- Ward K,
- Moore TH,
- Burke M,
- Davey Smith G,
- Casas JP,
- Ebrahim S
- Lichtenstein S,
- Slovic P
- 26.↵Her Majesty’s Government. Office for National Statistics. http://www.ons.gov.uk/ons/rel/regional-trends/regional-trends-online-tables/june-2011-release/topic-workbook-10–population-and-migration.xls. Accessed April 15, 2014.
- Finegold JA,
- Manisty CH,
- Goldacre B,
- Barron AJ,
- Francis DP
When we recommend that patients start primary prevention medications, we typically focus on risk as the deciding factor, with no discussion of aversion to taking medication. We rarely estimate the benefit from therapy in tangible terms. In this study, we produced tables of group-mean expectation of lifespan gain from taking a primary prevention therapy such as a statin. This is similar to a risk table, with the exception that younger patients show greater lifespan gain (despite lower short-term risk). We also surveyed 360 members of the general public, asking what level of lifespan gain would make it worth their while taking an imaginary tablet with ideal characteristics. Surprisingly, many demanded an expected lifespan gain larger than that available to any risk stratum of patients. Too little is known about how much patients dislike being on primary prevention medication or why this might be. Future research might explore this more formally. Ultimately, it may help to express to patients the size of the expected survival gain in a manner that can be easily understood. They, too, need a way to express the size of their dislike of being on preventative medication. Both could be expressed on a common scale, in terms of extra lifespan obtained or willing to be given up. If a patient expresses strong dislike of medication, it might trigger further discussion. Ultimately, these steps may help primary prevention becoming more truly personalized medicine.