Seth Seabury, PhD

To develop a model of the option value a therapy provides by enabling patients to live to see subsequent innovations and to apply the model to the case of nivolumab in renal cell carcinoma (RCC) and non—small cell lung cancer (NSCLC).

A model of the option value of nivolumab in RCC and NSCLC was developed and estimated.

Data from the Surveillance, Epidemiology, and End Results (SEER) cancer registry and published clinical trial results were used to estimate survival curves for metastatic cancer patients with RCC, squamous NSCLC, or nonsquamous NSCLC. To estimate the conventional value of nivolumab, survival with the pre-nivolumab standard of care was compared with survival with nivolumab assuming no future innovation. To estimate the option value of nivolumab, long-term survival trends in RCC and squamous and nonsquamous NSCLC were measured in SEER to forecast mortality improvements that nivolumab patients may live to see.

Compared with the previous standard of care, nivolumab extended life expectancy by 6.3 months in RCC, 7.5 months in squamous NSCLC, and 4.5 months in nonsquamous NSCLC, according to conventional methods. Accounting for expected future mortality trends, nivolumab patients are likely to gain an additional 1.2 months in RCC, 0.4 months in squamous NSCLC, and 0.5 months in nonsquamous NSCLC. These option values correspond to 18%, 5%, and 10% of the conventional value of nivolumab, respectively.

Option value is important when valuing therapies like nivolumab that extend life in a rapidly evolving area of care. 

Therapies that enable patients to live to see further innovations in care have option value. 

Option value raises the conventionally estimated value of nivolumab to patients with metastatic cancer by 18% in renal cell carcinoma, 5% in squamous non—small cell lung cancer (NSCLC), and 10% in nonsquamous NSCLC. 

Option value is particularly important in disease areas where there is rapid innovation. In such areas, payers and providers should consider option value when gauging the value of new therapies.

Cancer is a leading cause of death in the United States,

 with rising prevalence. In 1995, there were 7.1 million individuals in the US who had ever been diagnosed with cancer (2.7% of the population)

; by 2012, there were 13.4 million such individuals (4.3%).

 Although cancer mortality rates are declining,

 prevalence is expected to increase because of population aging,

 advances in cardiovascular care, earlier cancer detection, and life-extending treatment.

Increasing cancer prevalence translates to greater demand for oncology care, with implications for payers. Given growing pressures on payers’ budgets and debate over the cost of cancer treatment,

 value is an increasingly important consideration in coverage and access decisions. Moreover, the pressures on payers will increase as cancer prevalence rises, further reinforcing the importance of value.

As value has grown in importance in assessing cancer care, organizations have offered recommendations on measuring value. For example, the National Comprehensive Cancer Network and the American Society of Clinical Oncology recently published frameworks for considering a therapy’s efficacy and cost to determine its value.

 The Institute of Clinical and Economic Review, an independent health technology assessment (HTA) organization, uses cost-effectiveness analysis (CEA) and budget impact analysis to measure therapies’ value,

Traditional methods of CEA measure the value of an innovation by comparing benefits such as survival gains and improved quality of life with costs, assuming no other health technology improvements besides the one in question.

 However, such methods ignore aspects of therapies that matter to patients, like the value of hope, insurance value to healthy individuals, and the increased value people place on life near its end.

Such components of value can be difficult to quantify in HTA because they are subjective valuations based on unobserved patient preferences. However, 1 relevant component of value is rooted in real-world survival data, the “option value” of therapy. Option value is the benefit a therapy provides patients by enabling them to survive to the next innovation. For example, in cancer care, any survival gains today hold additional value because they increase the likelihood that a patient may live to access even more effective treatment in the future. This concept is particularly important in disease areas with traditionally poor outcomes and rapid innovation. 

Although metastatic non—small cell lung cancer (NSCLC) and metastatic renal cell carcinoma (RCC) have had low survival rates,

 innovation to treat both cancers is currently rapid. A number of breakthrough targeted therapies and immuno-oncology (IO) therapies, which harness the immune system to fight cancer, are available, such as nivolumab, pembrolizumab, osimertinib, and ceritinib for NSCLC and nivolumab and lenvatinib for RCC.

 These therapies are improving 1 or more clinical endpoints, such as survival rates for cancers with traditionally poor survival odds.

 Furthermore, the pipelines for both tumors are active.

Given the proliferation of treatment options and the prospect of future innovation, payers, policymakers, clinicians, and patients require multiple criteria to make care-related decisions.

 A therapy’s option value accounts for the future benefits of a life-extending therapy in addition to its immediate survival effects. With that in mind, we looked at the option value of nivolumab, the first IO therapy given breakthrough status and approved by the FDA for NSCLC and RCC.

 NSCLC is histologically divided into squamous and nonsquamous (adenocarcinoma, large cell) tumor types,

 and individuals face different prognosis and treatment options depending on the histology. Squamous NSCLC (about 25%-30% of all lung cancers) is typically smoking-related, whereas nonsquamous NSCLC is the most common type of lung cancer in nonsmokers (although smokers are still at greater risk than nonsmokers).

 We examined squamous and nonsquamous NSCLC separately.

We adapted a framework for estimating a therapy’s option value

 to the contexts of metastatic NSCLC and metastatic RCC. In particular, we examined the option value of nivolumab, the first programmed cell death protein-1 (PD-1) inhibitor in both disease states. In order to estimate nivolumab’s option value, a survival curve for patients taking the pre-nivolumab standard of care was needed first. The pre-nivolumab survival curves were modified using mortality hazard ratios (HRs) from clinical trial publications to obtain a nivolumab survival curve in each setting. Forecasting methods were used to project likely future survival improvements patients with either NSCLC or RCC may live to see. The conventional survival gain from nivolumab was obtained by comparing life expectancy with nivolumab, assuming no future innovation, against life expectancy with the pre-nivolumab standard of care. The option value of nivolumab was obtained by comparing life expectancy with nivolumab under 2 scenarios: allowing for likely future survival gains versus assuming no future innovation. Finally, the option value estimates were converted to economic terms. Additional methodological detail is available in the 

Survival trends in metastatic NSCLC and metastatic RCC were estimated using the Surveillance, Epidemiology and End Results (SEER) cancer registry.

 SEER, which tracks cancer incidence and mortality using data reported by registries across the United States, currently covers about 30% of the US population.

Long-term all-cause mortality rates were taken from the Human Mortality Database (HMD).

 The HMD contains detailed population and mortality data for 37 countries, and currently provides US life tables for 1933 through 2013.

Survival curves from SEER were modified to reflect nivolumab patients’ survival using mortality HRs from clinical trial publications.

 Specifically, in both squamous and nonsquamous NSCLC, nivolumab was compared with docetaxel, and in RCC, nivolumab was compared with everolimus. These comparisons reflect those in the trials noted on the nivolumab FDA label.

 The comparator drugs were assumed to represent the pre-nivolumab standard of care. 

We identified 3 study populations: metastatic squamous NSCLC, metastatic nonsquamous NSCLC, and metastatic RCC. First, we identified patients with NSCLC in SEER by requiring a primary cancer site of “lung or bronchus” or “trachea.” We distinguished squamous from nonsquamous histology types using 

International Classification of Diseases for Oncology, Third Edition

) histology codes. For both histology types, we defined metastatic NSCLC as American Joint Committee on Cancer (AJCC) stage IV, IIIb, or III with no surgery. The latter category addressed the fact that not all patients have a substage listed (eg, IIIa vs IIIb), and a lack of surgery suggested the case was inoperable.

To identify the RCC population, we required a primary site of “kidney,” “renal pelvis,” or “ureter,” and identified the RCC histology using 

 histology codes. To limit to patients with RCC with metastatic disease, we selected the “distant” stage using “SEER historic stage A.” Although AJCC staging is commonly used in clinical practice for treatment decisions and in clinical trials for eligibility,

 it was not well documented for RCC in SEER before 2004. Because metastatic RCC is a smaller sample than NSCLC, preserving sample size was important. Moreover, in order to track longer-term trends, it was necessary to be able to identify patients before 2004. In general, SEER distant stage approximates AJCC stage IV. Therefore, we identified metastatic RCC using the SEER distant stage. 

We estimated the option value of nivolumab in the 3 populations by taking these 5 steps.

Step 1: estimate pre-nivolumab survival curves. 

To establish a pre-nivolumab baseline, we estimated survival curves using the SEER data. We identified metastatic squamous and nonsquamous NSCLC and metastatic RCC patients diagnosed between 2001 and 2010. Survival curves for each population were estimated parametrically using a log-normal distribution. Covariates included age at diagnosis, age squared, gender, race, ethnicity, marital status, tumor grade, and a quadratic time trend.

Our survival estimation was limited by the patients’ duration of follow-up in SEER. Although we preferred using patients with a more recent diagnosis due to the pace of innovation in cancer care, this limited the number of years of follow-up available. Rather than project survival for many years beyond the duration of follow-up in the SEER data, we assumed that the minority of patients surviving more than 20 years past diagnosis had the same mortality rates as the general population, using US life tables from the HMD.

Specifically, we calculated mortality rates in the first 20 years after diagnosis using the estimated survival curves from SEER and then appended the HMD mortality rates. This lifetime mortality table was used as an input in the Lee-Carter model described in Step 2. 

The SEER survival curves estimated in Step 1 showed a clear trend of improving survival over time in all 3 populations. Based on these trends, we used the Lee-Carter method

 to forecast survival improvements for each population. The Lee-Carter method involves modeling age-specific mortality rates as a function of a long-term trend in overall mortality improvements and an age-specific response to the overall trend. This method is widely used as a benchmark for long-term mortality forecasts, including by the US Census Bureau and the Social Security Administration.

Step 3: estimate survival with nivolumab.

 To obtain survival curves for patients taking nivolumab in each population, we began with the pre-nivolumab survival curves estimated in Step 1. Based on the FDA label, which calls for second- or later-line use in both the NSCLC and RCC indications,

 we assumed that nivolumab would be given as a second-line therapy starting 1 year after diagnosis. Following the clinical trial evidence,

 we assumed that nivolumab patients would experience diminished mortality risk compared with patients on the prior standard of care. Specifically, we applied the mortality HRs from nivolumab trial publications to the pre-nivolumab survival curves estimated in Step 1. We conservatively assumed that any mortality benefits of nivolumab would disappear 4 years after beginning nivolumab therapy.

Step 4: calculate option value in terms of life expectancy gains. 

Using the survival curves from Steps 1 and 3 and the survival forecasts from Step 2, we calculated the undiscounted life expectancies of patients in each population for several scenarios. First, we calculated life expectancy under the pre-nivolumab standard of care, assuming no future innovation; call this 

. Second, we calculated life expectancy with nivolumab, assuming no future innovation; call this 

. Finally, we calculated life expectancy with nivolumab allowing for future innovation based on the forecasts from Step 2; call this 

We calculated the conventional survival gain with nivolumab by comparing life expectancy with nivolumab to that with the pre-nivolumab standard of care. Specifically:

Conventional survival gain with nivolumab = 

To calculate the option value of nivolumab, we compared life expectancy with nivolumab assuming no further innovation to expected lifetime with nivolumab allowing for future innovation. Specifically:

Step 5: calculate option value in economic terms. 

Finally, we expressed the conventional and option values in economic terms by applying an economic value of a life year to the additional life years gained. We selected a mid-range value from the literature of $150,000 per life year

 and applied a discount rate of 3% to obtain present discounted values. The per-patient option value was multiplied by the estimated size of each patient population to obtain the total economic values in each population. Population sizes were estimated using several sources.

To test the robustness of the study results to the underlying assumptions, several sensitivity analyses were performed. First, we estimated an alternative parametric survival model (log logistic). This tested the sensitivity of the results to our choice of parametric model. Second, we estimated survival exclusively using the parametric model rather than taking HMD mortality rates after 20 years. This tested the effect of our use of the HMD data. Third, we assumed nivolumab would be taken in the first year after diagnosis, rather than the second. This reflected the fact that metastatic cancer can progress rapidly and, therefore, some patients could potentially reach second-line therapy in the first rather than second year from diagnosis. Fourth, we varied the duration post diagnosis that nivolumab would be effective, from 2 to 10 years. This addressed the limited duration of clinical trial data and the accompanying uncertainty about the duration of nivolumab’s effects. Fifth, we estimated nivolumab’s mortality benefits using the confidence intervals of the published mortality HRs in order to address uncertainty about the exact magnitude of nivolumab’s effects of survival. Sixth, we varied the discount rate from 0% to 6%. Seventh, we varied the value of a life year from $50,000 to $250,000.

 Last, we varied the stage definition used for the selection of the SEER RCC data, using AJCC stage IV instead of the historic “distant” stage. (We used the historic stage definition in our base case analysis because it was better populated and we wanted to test the sensitivity of our results to this choice.)

 presents descriptive statistics on the 3 study populations in SEER. The nonsquamous NSCLC population was the largest, at 210,419 individuals, followed by squamous NSCLC (49,194) and RCC (12,868). The RCC population was younger than the squamous and nonsquamous NSCLC populations, with average ages of 64.5, 69.0, and 68.0 years, respectively. A higher share of patients with nonsquamous NSCLC (44%) was female compared with squamous NSCLC (35%) and RCC (33%). Patients with RCC were more likely to be married (60%) compared with patients with squamous (52%) or nonsquamous (53%) NSCLC. Racial composition was similar, with whites composing over 80% of all 3 groups. Median survival was 8 months in RCC and 7 months in both squamous and nonsquamous NSCLC.

All 3 populations showed an increasing 1-year survival rate in the 2001 to 2010 SEER data (

). Projected survival in 2011 to 2060 according to the Lee-Carter method is shown via the dashed lines. Survival improved the most quickly in RCC (0.44% per year), followed by nonsquamous NSCLC (0.39%) and squamous NSCLC (0.27%).

Life Expectancy Estimates and Option Value

Life expectancy for patients taking the pre-nivolumab standard of care was estimated to be 18.2 months in RCC, 11.7 months in nonsquamous NSCLC, and 11.6 months in squamous NSCLC. Ignoring future innovations, patients taking nivolumab were expected to survive 24.5 months with RCC, 16.2 months with nonsquamous NSCLC, and 19.1 months with squamous NSCLC, implying gains of 6.3 months, 4.5 months, and 7.5 months, respectively. These gains are shown in blue in 

 and represent the conventional survival gains from nivolumab.

However, during the time that nivolumab patients’ survival was extended, we allowed for additional innovations to come to market. Incorporating expected innovation, nivolumab extends survival by an additional 1.2 months in RCC, 0.5 months in nonsquamous NSCLC, and 0.4 months in squamous NSCLC. These additional gains due to option value represent 18% of nivolumab’s conventionally estimated survival gain in RCC, 10% in nonsquamous NSCLC, and 5% in squamous NSCLC.

Valuing the per-patient survival gains over the full incident metastatic population with each tumor, we found a conventional economic value from nivolumab of $775 million in RCC and an option value of $105 million (

). Due to its higher incidence, economic values were larger in NSCLC. The conventional value of nivolumab was $2.6 billion in nonsquamous NSCLC and $2.0 billion in squamous NSCLC, while the option values were $203 million and $73 million, respectively. Although these valuations are based on a mid-range value of a quality-adjusted life year, it should be noted that debate exists about the best way to provide economic valuations of incremental changes in survival, such as those in this study.

Articles

Option value is the benefit a therapy provides patients by enabling them to survive to the next innovation.

The Option Value of Innovative Treatments for Non–Small Cell Lung Cancer and Renal Cell Carcinoma

To study the association between cost sharing for diabetes medications, adherence, hospitalization rates, and healthcare costs, with relationship to patient risk.

A retrospective claims analysis of data from 35 large, private, self-insured employers (2004 to 2012).

We examined outcomes for 92,410 patients aged 18 to 64 years with a type 2 diabetes (T2D) diagnosis who filled at least 1 T2D prescription. First, we examined the relationship between adherence, measured as the proportion of days covered, and cost sharing, measured as the out-of-pocket cost to purchase a pre-specified bundle of T2D prescriptions. We then examined the association between adherence and hospital days. Simulations showed the effect of increased cost sharing on adherence and inpatient utilization.

A $10 increase in out-of-pocket cost was associated with a 1.9% reduction in adherence (

 <.01). In turn, a 10% reduction in adherence was associated with a 15% increase in per-patient hospital days (0.17 days; 

<.01). For the average plan, switching from low to high cost sharing reduced per-patient medication costs by $242 and increased per-patient hospitalization costs by $342, for a net increase of $100 in plan costs. Increases in per-patient costs were greater for high-risk patients, such as those with heart failure ($1328).

 Increased cost sharing for T2D medication was associated with reductions in pharmacy costs, but higher total costs for patients with T2D. This problem is particularly acute for patients with 1 or more cardiovascular comorbidities. The results suggest that increased diabetes cost sharing may hamper efforts to lower the total cost of diabetes care.

	This study investigated the association between cost sharing for type 2 diabetes (T2D) medications and patient outcomes, including medication adherence, hospitalizations, and healthcare costs to payers. Increasing patient out-of-pocket costs decreased patient medication adherence, and poorer medication adherence resulted in increased days spent in the hospital. The long-term hospitalization costs associated with lower adherence were greater than the health plan’s prescription drug cost-savings, particularly for high-risk patients. Decreasing out-of-pocket expenses may actually decrease long-term T2D-related costs for payers. 

		Higher levels of cost sharing are associated with lower levels of adherence to T2D medications. 

		Reduced adherence to T2D medications is associated with an increased number of days spent in hospital. 

		Although higher cost sharing reduces T2D drug costs, it raises hospitalization costs and, overall, increases costs to the plan. 

		High cost sharing has particularly adverse consequences in high-risk populations, such as those with heart failure or prior myocardial infarction, in terms of patient health and plan costs.

The last decade has seen a proliferation of payer-driven efforts to contain rising healthcare costs. Prescription drugs have increasingly been subject to utilization management policies, including tiered formularies and patient cost sharing for branded products. However, a large body of literature shows that increased cost sharing can have unintended consequences of patient noncompliance to prescribed medication regimens,

 and may delay treatment initiation in patients newly diagnosed with chronic illness, including diabetes, hypertension, high cholesterol, and multiple sclerosis.

 This has prompted concern that increased cost sharing may worsen outcomes for patients.

If high cost sharing worsens patient medication adherence, it could lead to costly complications that partially or completely offset payers’ pharmacy savings. For example, a study of patients with high cholesterol suggested that higher cost sharing reduced drug adherence and worsened outcomes for patients at high risk of complications, subsequently increasing overall costs.

 However, the relationships between cost sharing, medication adherence, and outcomes are sensitive to drug class and patient condition.

 More disease-specific estimates on the implications of cost sharing for patient outcomes and plan-spending are needed.

 In particular, in the case of diabetes, there have been a number of innovations over the past 10 years, and more information is needed on whether the relationship between cost sharing, adherence, and outcomes has changed as a result.

Disruptions to therapy caused by high cost sharing can be particularly harmful for patients with diabetes. Successful blood glucose control in patients with diabetes substantially decreases the risk of numerous complications, including cardiovascular disease (CVD), end-stage renal disease, stroke, retinopathy, and death.

 Diabetes affects 29.1 million Americans, costs the US healthcare system $245 billion annually (2012 data), and is projected to increase dramatically over the next 40 years.

 Type 2 diabetes (T2D), which represents 90% to 95% of adult diabetes cases,

 is a progressive disease, and requires most patients to eventually use drug therapy to achieve blood glucose control.

 Although antihyperglycemic drugs can effectively control blood glucose levels, research indicates that these drugs are underutilized, with studies finding the percent adherent to T2D medications ranges from 31% to 87%, with sizable groups of nonadherent patients in many populations.

Given the launch of new classes of antihyperglycemics in a healthcare system increasingly concerned with budget constraints, our study updates past work on cost sharing in diabetes

 in light of the new landscape. Moreover, given that past studies have found that increased cost sharing for diabetes medications may lead to complications which offset the pharmacy cost savings,

 one might wonder whether payers have changed benefit designs such that these offsets are no longer present. To answer these questions, we measured associations between cost sharing of T2D-related medications, patient adherence, and total costs for health plans. To do this, we used retrospective data on medication use, healthcare utilization, and spending among privately insured patients with T2D to simulate the per-patient healthcare spending associated with low-cost ($10 co-payment) and high-cost ($50 co-payment) pharmacy plans.

A retrospective cohort design was used to analyze the associations among cost sharing, medication adherence, and per-patient all-cause hospital days. Combining these analyses, it was possible to link cost sharing to hospitalization and estimate the impact of cost sharing on plan costs. Additional detail can be found in the 

The study sample was assembled using a longitudinal database of medical and pharmacy claims from 2004 to 2012 from 35 self-insured employers. The data set was constructed at the person-year level to reflect a person’s experience in a specific health plan.

Patients with T2D were identified for the study based on 1 or more inpatient or ambulatory visit with a T2D diagnosis (

International Classification of Diseases, Ninth Revision, Clinical Modificatio

n diagnosis codes 250.x0 or 250.x2). To create a working-age sample and exclude patients also covered by Medicare—for whom the claim history may be incomplete&mdash;the sample was restricted to patients aged 18 to 64 years at baseline. To ensure sufficient follow-up for analysis, patients were required to have had continuous health plan and pharmacy benefit enrollment during the first observed year with a T2D diagnosis, and for at least 1 year afterward.

Patients excluded from the study included those with less than 1 year of follow-up, those pregnant, and those with gestational diabetes. Additionally, patients using an insulin pump were excluded due to the difficulty in measuring adherence in claims data; however, patients using insulin, not on a pump device, were included. Because this study examined adherence to T2D medications, person-year data were retained only for years in which the patient filled 1 or more T2D prescriptions. Data from the first year with a T2D diagnosis were excluded because the patient may not have had T2D for the entire year. Finally, we excluded outlier patient data in terms of cost sharing and hospital days, as well as patients in small plans (<10 total enrollees with T2D). Additional details on outlier exclusions are available in the eAppendix.

Data were collected for the following covariates: patient age, gender, indicators for each of the Charlson comorbidities

 during the previous year, medical cost sharing faced by the patient in the given year, and indicators for each year in the study period.

 Medical cost sharing was defined as the average portion of medical (ie, inpatient, outpatient, and emergency department) services that individuals in a given plan paid out of pocket (OOP).

The key explanatory variable was cost sharing for T2D medications, measured as the average patient OOP cost of purchasing a T2D drug prescription in a given plan. Specifically, the OOP costs to obtain a fixed bundle of T2D prescriptions were averaged across the individuals in a plan, and the result was divided by the number of drugs in the bundle to obtain the average OOP cost per T2D prescription. Total T2D drug costs were defined as the total amount the plan and patient spent on T2D drugs in a given year.

The key outcome measures were patient medication adherence and number of hospital days. Adherence was measured as the proportion of days covered (PDC) (ie, the fraction of days in the year in which the patient had a supply of at least 1 T2D medication on hand). For example, a PDC of 0.75 indicates that a patient had T2D medications on hand for about 9 months of the year. PDC was selected over another common adherence measure—the medication possession ratio&mdash;which has a tendency to overestimate adherence. Hospital days were defined as the number of days an individual spent in hospital in a given year. Hospital stays in which a patient was admitted and discharged in a single day were counted as a full day.

Linear regression models were used to estimate the relationship between cost sharing, adherence and hospital days at the patient-year level. First, adherence was estimated as a function of drug cost sharing, then days in hospital were estimated as a function of adherence. Adjusted analyses controlled for age, gender, each of the Charlson comorbidities, and the year (to absorb time trends). Because patient adherence to T2D medications may also depend on OOP spending for other healthcare services, the analysis of adherence, as a function of T2D medication cost sharing, also controlled for medical cost sharing.

We used these estimates in a 2-step process to predict total plan costs (ie, plans’ drug plus hospitalization costs) at high and low levels of T2D drug cost sharing. Low cost sharing and high cost sharing were defined as an average OOP cost per T2D prescription of $10 and $50, which approximately corresponded to the 10th and 90th percentiles of cost sharing, respectively. Regression models were used to predict adherence and hospital days at the low and high cost-sharing levels, holding covariates at their mean values. Payer hospitalization cost was determined by multiplying the predicted number of days in hospital by the average payer cost of a hospital day. Costs were inflated to 2012 US dollars using the Consumer Price Index for All Urban Consumers. By combining the analyses linking cost sharing to adherence and adherence to days in hospital, we predicted the plan’s T2D drug costs and hospital costs at high and low cost-sharing levels, and estimated how total plan cost would vary with cost sharing.

These predictions were performed for several scenarios. The base case used the full study sample. Subsequent analyses were performed on subsamples defined by patient risk: specifically, the presence of CVD, congestive heart failure (CHF), prior myocardial infarction (MI), renal disease, or prior hospitalization. Additional subsamples included lower-risk patients, including those in none of the above risk groups, and those with no comorbidities. Finally, because our plan-level cost-sharing measure was potentially subject to measurement error in smaller plans, the analyses were repeated based on plan size subsamples. Our base-case analysis included all T2D patients in plans with 10 or more patients with T2D that year; subanalyses also examined plans with greater than 20, 50, 100, and 500 patients with T2D.

From more than 7 million covered lives in the database, 727,476 had at least 1 claim with a T2D diagnosis. After applying exclusion criteria, 92,410 patients from 1514 healthcare plans were eligible for study inclusion. Plans had an average of 8776 members and 144 patients with T2D in the study cohort. Table 1 shows the personal characteristics of patients included in the study overall, as well as those in low and high cost-sharing plans (defined by the 10th and 90th percentiles of cost sharing). The sample was mostly male (56.1%), aged between 50 to 64 years (71.4%), and had an average of 1.17 Charlson comorbidities at baseline.

The mean annual per patient OOP amount for T2D drugs was $242 (SD = $337) and the mean OOP amount per individual T2D prescription was $22.70 (SD = $16.70). Only 10% of patients had an average OOP per T2D prescription cost exceeding $54, and 10% had an average OOP cost below $11.

The average adherence to T2D medications, measured as PDC, was 67% (SD = 29.1%). Patients with T2D spent an average of 1.15 (SD = 7.70) days in hospital each year. In a given year, 1 in 6 (16.6%) patients with T2D was hospitalized. Adherence in the sample ranged from 0.27% to 100%. (No one had a zero PDC because individuals were required to fill ≥1 prescription for cohort inclusion.) Individuals in low cost-sharing plans had an average adherence of 69%, whereas in high cost-sharing plans, the average adherence was 63%.

Compared with their counterparts in low cost-sharing plans, individuals in high cost-sharing plans generally tended to be younger (average age of 50.8 vs 53.2 years) and healthier (eg, fewer comorbidities, fewer days in hospital), consistent with past research.

 This suggests that any negative effects of cost sharing could be understated in our analysis due to selection of healthier patients to less generous plans.

Analysis Figure 1 illustrates the negative association between cost sharing and adherence for T2D prescription drug therapy. The figure, which reports the unadjusted average co-payment per prescription to the average medication adherence per plan, shows that a $10 increase in the OOP cost of T2D drugs was associated with 1.6 percentage points lower medication adherence (

 <.01). The average PDC in the plan was 67%, which means a $10 increase in OOP costs is associated with a —2.4% change in adherence (or 1.6 percentage points out of 67).

 Although Figure 1 shows a negative relationship between adherence and cost sharing, it does not account for differences across patients within the plan. After adjusting for observed patient characteristics, the negative relationship between cost sharing for T2D prescriptions and adherence to T2D drug therapy persisted. Based on the regression, a $10 increase in the OOP cost of T2D drugs was associated with a lower PDC of 1.3 percentage points (

 <.01), or a —1.94% change in drug adherence for a plan with the average PDC of 67% (or 1.3 percentage points out of 67).

 Increased adherence was associated with reduced all-cause hospital days. Each 10% point increase in adherence was associated with a 0.17-day reduction in the number of days a patient was expected to spend in hospital annually (

 <.01) (Figure 2). For a plan with 100,000 members with T2D, this equates to a reduction of 17,000 patient-days in hospital annually. This represented a 14.8% decrease in the mean 1.15 hospital days per patient, per year.

Increased cost sharing for T2D drugs was associated with increased payer hospitalization costs. Raising the average OOP cost per T2D prescription from $10 to $50 was associated with a $242 reduction in per-person payer costs for T2D prescriptions and an increase of $342 per person of payer hospitalization costs, as shown in Figure 3.

Impact of increased cost sharing by risk group and plan size. 

As shown in Table 2, increased cost sharing was associated with higher total plan costs across a variety of plan sizes and patient populations. Increasing the average OOP payment per T2D prescription from $10 to $50 resulted in increased overall payer costs (combined drug and hospitalization costs) for all high-risk and 1 of 2 low-risk patient populations. The largest increase in plan costs was for patients with T2D and CHF ($1328 per patient, per year). The only patient risk group for which increased cost sharing did not increase overall plan costs was patients with no comorbidities other than diabetes.

Although plans of all sizes experienced increased overall payer costs, the plan impact of increased cost sharing was highest in the largest plans. Among plans with more than 500 members with T2D, increased cost sharing was associated with a $169 increase in per-person plan costs compared with $100 in plans with 10 or more members with T2D. This means that for a plan with 1 million covered lives and 150,000 members with T2D (assuming 15% prevalence), increasing cost sharing would raise plan costs by $25 million annually.

Efforts to contain healthcare costs have led payers to shift the disease management burden to patients through increased prescription drug cost sharing. In this study, we estimated how increased cost sharing affected medication adherence, hospital days, and total per-patient health plan costs for patients with T2D using claims data. We found that each $10 increase in patient OOP cost was associated with a 1.6 percentage point decrease in adherence (2.4%; 

 <.01); moreover, a 10% reduction in adherence was associated with 15% higher per-patient hospital days. Combining these results, we found that for the average plan, switching from low to high cost sharing would potentially reduce plan per-patient medication costs by $242 and increase plan per-patient hospitalization costs by $342, for a net increase in plan cost of $100. Thus, although higher cost sharing reduces T2D drug costs to the plan, it may raise hospitalization costs by a greater amount, more than offsetting the savings. For a plan with 150,000 members with T2D, this could amount to an increase of $25 million in annual plan costs.

High cost sharing has particularly adverse consequences for patient outcomes and plan costs in certain high-risk populations (eg, patients with various comorbidities and/or prior hospitalizations) in our model. In particular, high cost sharing among patients with CHF, CVD, and prior MI led to simulated large increases in per-patient, annual hospitalization costs, at $1794, $1220, and $632, respectively. These costs far outweighed any per-person savings associated with medication cost sharing, at $466, $272, and $149, respectively. Likewise, healthcare plans with more than 500 patients with T2D experienced the highest increase in per-person annual hospitalization costs, at $550—which was higher than per-person cost savings associated with medication cost sharing, at $381. On the other hand, for the healthiest patients with no comorbidities during the study period, cost sharing on diabetes medication resulted in modest cost savings. However, the relationship between cost sharing and adherence was consistent across the high-risk and low-risk groups, so it is unclear whether these savings would persist over longer time periods if patients failed to adhere to their medications and their diabetes worsened or led to complications. This should be considered in future work.

These findings are consistent with prior work demonstrating that cost sharing worsens patient medication adherence, resulting in costly complications. For example, previous studies have shown that patient cost sharing leads to decreased medication adherence and can compromise health in patients with T2D and other chronic conditions.

 Results such as these led the Congressional Budget Office to change its scoring system, such that every 10% increase in prescription medication use reduces estimated medical costs by 2%.

 This study confirms these findings and further demonstrates that decreasing cost sharing for patients with T2D may be advantageous for payers. Specifically, lower patient OOP expenses for T2D medications may improve access and adherence, leading to improved long-term patient outcomes and decreased overall costs.

 On the other hand, increased medication cost sharing may decrease adherence and associated total payer costs.

These findings demonstrate the importance of a comprehensive approach to benefit design incorporating costs, value, and patient outcomes. As T2D prevalence increases in the United States, the cost burden of T2D care will increase for payers,

 stimulating further efficiencies in healthcare delivery systems. As plans experiment with cost sharing and tiered prescription plan designs to lower payer costs, some patients may face substantial, and sometimes prohibitive, OOP costs.

 Plans may also sort newer T2D therapies into higher cost-sharing tiers, with possible consequences for drug initiation and adherence, as well as overall plan costs.

 Our findings suggest that, rather than imposing policies that limit or disrupt T2D care, payers may see greater benefit from policies that carefully consider cost sharing and adherence promotion. More generally, formulary policy, like all aspects of healthcare delivery, should be designed to encourage the use of services that foster quality, improve outcomes, and offer the most value to patients and society.

An analysis of claims from over 90,000 patients with type 2 diabetes (T2D) demonstrates that increased medication cost sharing is associated with higher rates of hospitalization and increased plan costs.

Seth Seabury, PhD

Articles

The Option Value of Innovative Treatments for Non–Small Cell Lung Cancer and Renal Cell Carcinoma

Impact of Type 2 Diabetes Medication Cost Sharing on Patient Outcomes and Health Plan Costs