When Is It Cost-effective to Change the Behavior of Health Professionals?

Economic analyses commonly estimate the incremental costs and benefits of using different treatments on the basis of the best available clinical evidence. They do not include the costs of achieving an appropriate adoption of results by physicians. Important research findings often do not translate automatically into practice^{1 - 3} while new and well-marketed products may sometimes achieve a greater market share than merited by their additional costs and benefits.⁴ Consequently, policymakers may have to commit additional resources to influence physician behavior if desirable changes in practice are to be achieved. We provide a framework for exploring the economics of influencing physician behavior. Using 2 examples, our purpose is to distinguish between treatment cost-effectiveness (the incremental costs and benefits of a treatment) and policy cost-effectiveness (combining treatment cost-effectiveness with the cost and magnitude of change achieved by an implementation method).

In England, there is evidence that angiotensin-converting enzyme (ACE) inhibitors are underused in the care of patients with heart failure,⁵ while newer classes of antidepressants have achieved widespread first-line use without demonstrating added value.⁴ Clinical practice guideline development groups, having studied the profile of costs, benefits, and harms for these conditions, have recommended that physicians should address these shortcomings. The recommendations feature different economic messages. Every patient with depression treated first-line with a selective serotonin reuptake inhibitor (SSRI) instead of a tricyclic antidepressant costs an additional $75 per episode without demonstrable clinical benefit.⁶ Patients with heart failure not treated with an ACE inhibitor are denied a cost-effective treatment,⁷ valued at $2156 per life-year gained.⁸ Costings, originally conducted in pounds sterling, have been converted to US dollars (£1 = $1.50) for ease of exposition.

A health policymaker may suspect many instances of suboptimal care but what, and how much, can be done about it? The answer depends upon a number of steps (Figure 1). Implementation requires a clear and deliverable evidence-based message, evidence that current care is suboptimal and the message is not being applied, a robust estimate of the cost and impact of alternative methods of behavioral change, and understanding of the local organization of health care.⁹ With information at each step, a policymaker is able to understand the likely costs and health consequences of intervening in instances of suboptimal care.

The costs and benefits of influencing physician behavior are derived in the Article . This presentation shows that having to invest resources to change physician behavior imposes an addition, or loading factor, upon treatment cost-effectiveness. It is transparent that policy cost-effectiveness is likely to remain attractive in those treatments that are highly cost-effective and likely to become unattractive when cost-effectiveness of treatment is borderline at the outset. In general, whether implementation by any given method is worthwhile involves a complex interplay of factors. Cheaper implementation methods achieving greater levels of change reduce the loading effect. Similarly, larger health gains per patient, higher prevalence of disease, larger practice size, or longer duration of behavioral change all reduce the loading, all other things being equal. When the loading is small, treatment and policy cost-effectiveness are very similar. When the loading is large, a cost-effective treatment may not be worthwhile pursuing as a policy goal, using behavioral change methods.

Summary of Evidence for Treatment (Cost-effectiveness)

Δc_t = net cost of care per additional patient treated

Δb_t = health gain per additional patient treated

ΔCE_t = treatment cost-effectiveness, Δc_t /Δb_t

Summary of Evidence for Implementation

Δc_i = net cost of implementing change per practice

Δb_i = change in proportion of patient care following guidance

ΔCE_i = implementation cost-effectiveness, Δc_i /Δb_i

Local Implementation Scale Factors

n = the number of practices (or hospitals) involved

N = the number of patients targeted

Net policy cost: ΔC_p = Δc_i × n + Δb_i × N × Δc_t

Net policy benefit: ΔB_p = Δb_i × N × Δb_t

Policy Decision (Cost-effectiveness)

n_p = the average practice size

p_d = population prevalence of the condition targeted

d = duration of effect

Policy Decision (Cost-minimization)

Policy cost: ΔC_p = ΔC_i + ΔC_t = Δc_i×n + Δb_i × N× Δc_t

Some treatment decisions involve selection between alternatives with equivalent outcomes. The policy decision to influence behavior then centers on cost alone. The decision rule simplifies to one where the overall cost-saving from changed patterns of health care must exceed the overall cost of implementation.

Educational outreach visits by pharmacists to primary care physicians is one method to promote changes in prescribing medications. A recent randomized controlled trial of Evidence-Based OutReach (EBOR) explored the impact of this method, incorporating recommendations for the use of ACE inhibitors and antidepressants.^{10 - 11} Data on 11 328 patients were collected from 162 general practitioners in 69 practices.

In England, the National Health Service is divided for administration purposes into local Health Authorities with an average population of about half a million. Primary health care is provided by general practitioners who organize themselves into practices of varying size, but with an average of 3 to 4 practitioners and about 6000 registered patients. Overall, outreach had a significant effect measured as the change in adherence to guideline recommendations across the guideline topics (odds ratio [OR], 1.2; 95% confidence interval (CI), 1.1-1.4). Applying this OR to a baseline adherence rate of 40%, there is a 5.2% (95% CI, 1.7%-8.7%) increase in the number of patients treated according to the recommendations. Small practices (with 1 or 2 practitioners) demonstrated a substantially greater response to outreach than larger practices. Consequently, adherence in smaller practices improved by 13.5% (95% CI, 6.0%-20.9%) from a baseline adherence rate of 40%.

Based on the use of resources in the EBOR trial, a similar willingness to receive outreach, and absence of scale effects, outreach for 1 guideline is estimated to cost $38 400 per Health Authority (Table 1).

The policy cost-effectiveness for the 2 examples using educational outreach is summarized in Table 2. The cost per guideline per practice per unit change in behavior (ΔCE_i) is $11 690 (95% CI, $6980-$35 740). The effect of the implementation loading ($446/life-year gained) upon the treatment cost-effectiveness of ACE inhibitors ($2156/life-year gained) for heart failure is small and a policy of implementation is cost-effective ($2602/life-year gained). Academic outreach to promote a reduction in use of SSRIs in favor of tricyclic antidepressants is not estimated to be cost-saving: the cost per patient of outreach ($82) exceeds the cost-saving from behavioral change ($75). This would not be worth pursuing by educational outreach.

The impact of educational outreach on small practices (1 or 2 partners) is modeled as a subgroup analysis (Table 2). Although an increase in outreach effectiveness (Δb_i) reduces the loading factor, this is partially offset by a reduction in numbers of patients per practice targeted by each guideline. Nonetheless, intervention to reduce the use of SSRIs has become cost-saving, although the magnitude of saving is imprecise.

As with any model, a number of assumptions are necessary. Local patients are assumed typical of those enrolled in trials and are treated to the same extent. That is, patients treated are being counted rather than the amount of treatment received by each patient. The findings of the implementation studies may have to be assumed transferable not just to different localities but with different evidence-based messages and diseases. As with the costs of treatments, the costs of outreach may be anticipated to vary by locality and country. Implementation studies that report units of component resources, disaggregated from their unit costs, will help users derive a valid local cost of implementation by applying local patterns of resource use and prices. Economies of scale may be important since each guideline has fixed training costs but may be implemented on a varying number of practices.

The effect of time on the magnitude of behavioral change achieved by implementation methods is uncertain. It is unlikely that change is constant in magnitude or permanent. A recent evaluation of a computer-based clinical decision support system to prevent venous thromboembolism highlighted the need for continuing implementation of the system to achieve lasting behavioral change.¹⁴ In the EBOR trial, measures of adherence to guidelines were made at baseline and after 12 months with intervention occurring between 3 and 6 months. For simplicity it has been assumed that outreach has an effect for 1 year and then ceases, at which time further outreach activity is required. This is analogous to assuming a full effect for the first 6 months reduced by 50% every subsequent 6 months.

The perspective in our illustration is a local policymaker faced with the decision to implement national guidelines. The principal costs associated with outreach visits are for the salaries of the outreach workers and their facilitators (Table 1). There are also costs associated with the recruitment of outreach workers, training, and the development and production of local promotional materials. In these examples, the cost of guideline development has been explicitly excluded, because in England and Wales, there is a commitment to produce evidence-based guidelines and related audit measures; from the perspective of the analysis, the guidelines are free goods. Similarly, physicians are expected to undergo continuing education and audit; time participating in these activities has not been charged. Other assumptions are possible as necessary for particular situations.

Health systems experience considerable pressure to provide new health technologies, sometimes of borderline cost-effectiveness, while long-established, cheap, and effective but unfashionable health care may not be used. The model we present shows that multiple influences determine whether investing to achieve behavioral change is worthwhile. The preconditions are that there is an evidence-based message of acceptable clinical and cost-effective care, and current care is suboptimal. Worthwhile behavioral change from a policy viewpoint requires an implementation method that does not load treatment cost-effectiveness to such an extent that normal bounds of cost-effectiveness are exceeded.

In the main analysis, the cost-effectiveness message (ACE inhibitors) was worth implementing but the cost-saving message (antidepressant selection) was not. The large health benefit achieved by ACE inhibitors used in heart failure favorably influenced both treatment cost-effectiveness and the loading term. Changing the choice of antidepressant for depression achieves small cost-savings that do not exceed the cost of the implementation method. The promotion of interventions to reduce mortality or major morbidity is more likely to prove attractive than those switching prescriptions from a more expensive to a less costly pharmaceutical. The relatively high cost of influencing prescriptions can quickly overwhelm potential cost-savings.

We have presented a method to help decision makers explore whether investing in change is worthwhile. This may be applied in any currency and for any unit of outcome, and is valid when investing to reduce as well as increase activity. Although educational outreach visits to primary care physicians by pharmacists are a popular strategy, their impact is modest. As more implementation strategies are evaluated for their cost and level of behavorial change, it should be possible for any condition with known prevalence to estimate loading adjustments for treatment cost-effectiveness, and thus to determine the best implementation method to meet policy goals.