 Due to volatility in healthcare costs, shared savings contracts can create systematic financial losses for payers, especially when contracting with smaller providers. To improve the business case for shared savings, we calculated the prices of financial options that payers can “sell” to providers to offset these losses.

Using 2009 to 2010 member-level total cost of care data from a large commercial health plan, we calculated option prices by applying a bootstrap simulation procedure. We repeated these simulations for providers of sizes ranging from 500 to 60,000 patients and for shared savings contracts with and without key design features (minimum savings thresholds,bonus caps, cost outlier truncation, and downside risk) and under assumptions of zero, 1%, and 2% real cost reductions due to the shared savings contracts.

Assuming no real cost reduction and a 50% shared savings rate, per patient option prices ranged from $225 (3.1% of overall costs) for 500-patient providers to $23 (0.3%) for 60,000-patient providers. Introducing minimum savings thresholds, bonus caps, cost outlier truncation, and downside risk reduced these option prices. Option prices were highly sensitive to the magnitude of real cost reductions. If shared savings contracts cause 2% reductions in total costs, option prices fall to zero for all but the smallest providers.

Calculating the prices of financial options that protect payers and providers from downside risk can inject flexibility into shared savings contracts, extend such contracts to smaller providers, and clarify the tradeoffs between different contract designs, potentially speeding the dissemination of shared savings.

Volatile healthcare costs and asymmetric shared savings contracts can create systematic financial losses for payers.

		Calculating the fair prices of “options” that payers can sell to providers to offset these losses can improve the business case for shared savings and clarify tradeoffs between contract designs.

		In a large sample of commercial health plan members, simulated option prices reached 3.1% of overall healthcare costs for providers with just 500 members. Prices were considerably lower for larger providers and more balanced contract designs.

		If shared savings contracts create 2% real savings, selling options may be unnecessary for most providers.

Shared savings contracts between payers and providers are important features of many innovative healthcare payment and delivery models, including Accountable Care Organizations (ACOs) and medical homes.

 These contracts incentivize cost  containment by allowing providers to share in any savings they create, usually by paying them a percentage of the difference between observed and expected costs for their patient panels. However, providers generally do not pay an equivalent penalty if observed costs exceed expectations. This asymmetric risk distinguishes shared savings contracts from pure capitation or “global payment” arrangements in which providers stand to absorb gains and losses in equal measure.

For a given patient, costs of care from year to year exhibit volatility, or unpredictable variation due to chance alone. Patients who are relatively healthy with low health expenditures in one year may become sick with high health expenditures in the next, and vice versa. Due to this random variation, observed costs for a provider’s patient panel can vary widely from expected costs, producing apparent savings in one year and cost overruns in the next. Therefore a substantial proportion of shared savings bonuses may be paid due to chance alone (“paying for volatility”), even if providers do nothing to contain costs. While random variation also can produce cost overruns, the resulting penalties will not counterbalance random rewards when shared savings contracts are asymmetric.

Faced with the prospect of paying for volatility, payers justifiably may avoid shared savings contracts: unless payers are confident  that shared savings contracts will result in highly effective cost-containment efforts, random cost variation can undermine payers’ business case for such arrangements.

 Whether due to lack of confidence in cost-containment ability or aversion to financial risk, providers have resisted contracts that include penalties for cost overruns.

 Therefore to enable widespread adoption of shared savings contracts, we need tools to bridge the gap between providers who want protection against downside risk and payers who need a winning (or at least neutral) business proposition.

Tools to improve payers’ business case for shared savings contracts work in 2 general ways: reduce the amount of random variation (ie, volatility) in costs, and make the contract more symmetric. To reduce random variation in costs, payers frequently require patient populations larger than a specified minimum threshold.

 For example, Medicare ACO demonstrations require 5000 or more patients (for the Shared Savings Program) and 15,000 for non-rural Pioneer ACOs (

 Due to the “law of large numbers,” larger patient populations have less random variation as a percentage of overall costs, thereby reducing payers’ risk of losses.

Truncating cost outliers (patients with extremely high costs) also can reduce random variation in shared savings contracts. The Medicare Shared Savings Program truncates costs at the 99th percentile: for the purposes of calculating savings, the most expensive 1% of patients are considered to have costs equal to the patient at the 99th percentile.

 Both expected and observed costs are truncated, so while this method reduces the threat that cost outliers pose to providers, it also produces lower expected costs (ie, a tougher target for providers to beat).

To make shared savings more symmetric and improve the likelihood that bonuses are paid for real savings rather than volatility, some contracts include provisions that reduce the shared percentage of observed savings. For example, Medicare Pioneer ACO and Shared Savings Program contracts include both minimum savings rates that must be achieved before providers receive bonuses (ranging from 1 to 3.9%, depending on ACO size) and maximum caps on the amount of bonus any provider can receive (ranging from 10 to 15% of expected costs in the Medicare ACO programs).

Penalizing providers for cost overruns also makes shared savings contracts more symmetric. However, even in programs with downside risk for providers, some asymmetry generally persists. For example, when quality criteria are met the Medicare Pioneer ACO and Track 2 of the Medicare Shared Savings Program penalize providers 40% of the amount of cost overruns—less than the 60 to 70% bonuses triggered by savings.

By limiting shared savings contracts to large providers, payers may miss opportunities to engage ambitious smaller ones (like primary care practices transforming into medical homes) in cost containment. In addition, the relative strengths and weaknesses of current solutions can be unclear. For example, how does excluding outliers compare with requiring a minimum savings rate before sharing begins? Is the answer the same for small primary care practices and large ACOs? What about combinations of these features? We believe a new tool can help evaluate a wide variety of shared savings arrangements and enable such contracts with small providers.

To allow a broad range of providers to enter asymmetric shared savings contracts without systematically disadvantaging payers, we propose to borrow a commonly used financial tool: the “call option.”

 When an investor wants to profit if a stock price climbs but not incur a loss if the price falls, he or she can purchase a call option that protects against downside risk. These options are not cost-free to the investor; instead, investors must pay for them.

Shared savings contracts present an analogous situation, and they might spread more rapidly if payers offer to sell “call options” to providers as a condition of entering such contracts. These options can be priced to counterbalance exactly payers’ expected losses due to random variation in healthcare costs, under any shared savings arrangement, with providers large and small. Once calculated, these option prices can clarify the business cases for both payers and providers, enabling construction of customized shared savings contracts when both parties are willing to negotiate.

Using total cost data from a large sample of members in a commercial health plan, we simulated differences between observed and expected costs due to chance, calculated the resulting shared savings bonus payments (ie, payments for volatility), estimated the average financial loss to payers causedby these bonus payments, and priced a call option to offset these expected losses. We then extended the simulation to scenarios where shared savings contracts caused “real savings” (ie, reductions in costs due to providers changing their care).

We obtained data on total healthcare spending in 2009 and 2010 for each continuously enrolled member of a large commercial health plan who was attributable to a pilot or control practice in a current medical home demonstration. By selecting patients attributable to primary care providers based on office visits, our patient sample consisted of those who might be included in shared savings contracts (and not those who would be impossible to attribute to a specific provider).

 To ensure that our simulations were not affected by the medical home demonstration, we excluded all patients attributed to pilot practices.

These spending data represented the total allowed amounts for all medical and prescription drug claims, including both the plan’s expenditures and any deductibles or coinsurance amounts paid by patients. Because we wanted to compare the financial impact of shared savings arrangements with fee-for-service payment, we excluded members of health plan products that featured other types of payment (eg, capitation). After exclusions, our study population comprised 77,364 patients.

To simulate shared savings contracts, we first calculated the expected costs for each patient in 2010 based on his or her 2009 costs. To do this, we assumed perfect prediction of general medical inflation for the entire patient population. Our assumption of a population-level cost prediction model that cannot be improved allowed us to focus cleanly on how volatility in patient-level costs affected shared savings contracts.

Of the Medicare ACO programs, our method for calculated expected costs is most similar to the Pioneer demonstration, which also performs risk adjustment using same-patient historical expenditures.

 However, because commercially insured populations tend to be younger and healthier than Medicare beneficiaries (among other differences between these populations), we did not attempt to replicate exactly any Medicare ACO contracts.

We simulated a variety of shared saving contracts, each defined by 5 key dimensions (

): percentage of savings shared; minimum savings rate, with and without first-dollar sharing; shared savings bonus cap; cost outlier truncation; and provider penalties for cost overruns. The main analyses varied these dimensions 1 at a time, but in supplementary analyses we explored their combinations. Although shared savings contracts can be limited to just 1 kind of spending (eg, just outpatient care),

our scenarios included comprehensive costs of care across all providers and types of services.

Simulating Shared Savings and Calculating Option Prices

To estimate the financial impact of shared savings contracts, we performed a bootstrap simulation by randomly drawing samples of patients from the study population. We simulated providers of many sizes by drawing samples ranging in size from 500 to 60,000 patients, with each sample representing a provider’s patient panel (or a “simulated provider,” for short).

 We included panel sizes smaller than 2000 (a size that might be expected for a solo primary care provider) because in general, a single health plan rarely will enroll a provider’s entire patient panel. For example, a shared savings contract between a solo practitioner and a plan enrolling 25% of his or her overall panel might include only 500 patients. Within each size category, we repeated this process 10,000 times, effectively generating 10,000 simulated providers.

For each simulated provider, we calculated total costs per patient in 2009 and 2010. Within each provider size category, we used the average ratio of 2009 to 2010 actual costs per patient (across all 10,000 simulated providers) to calculate an expected 2010 total cost for each simulated provider. We then calculated, for each simulated provider, the difference between actual and expected 2010 costs. If negative, this difference was “savings;” if positive, it was a “cost overrun.”

This process preserved year-to-year random variation in costs at the provider level while guaranteeing perfect population-level cost prediction. In other words, the average savings across all providers was assumed to be zero in the base case, even though individual providers could have large simulated savings or cost overruns due to fluctuations in the costs for their specific patient panels. Because all patient samples were chosen at random, any difference between actual and predicted costs could only be due to chance—not providers altering their care.

For each simulated provider, we calculated the amount of shared savings (or penalty if a cost overrun was present) according to the terms of each shared savings contract scenario. We then calculated the average per patient financial impact to the plan within each provider size category. If the financial impact was negative, the magnitude of this impact was equal to the fair price of a call option that would eliminate the plan’s average loss due to the shared savings contract.

Because shared savings contracts are designed to incentivize providers to generate cost savings, we repeated the above calculations assuming that the shared savings contracts caused providers to successfully alter their care, yielding 1% or 2% reductions in average 2010 per-patient costs.

Data management, simulations, and analyses were conducted using SAS software, version 9.2 (SAS Institute, Cary, North Carolina). This research was approved by the RAND Human Subjects Protection Committee.

In the study population, mean costs of care were $6406 in 2009 and $7166 in 2010 (

 the distribution of healthcare costs in the population was highly skewed, with the costliest 20% of patients accounting for approximately 80% of total expenditures in each year. Patient-level costs were also volatile; only 50% of the patients in the most costly quintile in 2009 remained in this quintile in 2010.

As expected, simulated option prices were greater for smaller providers than for larger providers (

). In the shared savings “base case” (50% sharing of any savings, with no penalty for cost overruns), the fair prices of options were $225 per patient (3.1% of total expected costs) for providers with 500 patients, compared with $23 per patient (0.3%) for providers with 60,000 patients. Therefore, to turn a profit, a provider having only 500 patients would need to achieve an observed 6.2% savings rate, the minimum necessary for the shared savings bonus to offset the option price. However, even without altering their care after purchasing the option, 31% of such providers would be expected to profit due to chance alone (ie, the magnitude of their bonus would exceed the option price). From the payer’s perspective, these randomly generated provider profits would be offset, on average, by the options purchased by less lucky providers of the same size.

A 2% minimum savings rate that featured first-dollar sharing had little effect on option prices for smaller providers, decreasing the price for 500-patient providers from $225 to $223 per patient. Even for 10,000-patient providers, the 2% minimum savings rate reduced the option price from $57 to $50 per patient, with 28% of such providers profiting due to chance alone. However, this minimum savings rate halved the option price for 60,000-patient providers and reduced the rate of profits due to chance from 33% to 11% for these providers. Without first-dollar sharing, a 2% minimum savings rate produced lower option prices in all provider size categories.

A 10% bonus cap produced greater reductions in option prices for smaller providers ($225 to $193 for 500-patient providers) than for larger providers. Outlier truncation at the 99th percentile of patients reduced option prices by approximately 50% for providers in all size categories and slightly increased the percentage of providers who would profit due to chance alone (with the magnitude of these profits being lower than without truncation [data not shown]).

Articles

WEB EXCLUSIVE

By pricing options that protect providers from downside risk,health plans can more clearly evaluate different shared savings contracts and expand them to smaller providers.

Option Pricing: A Flexible Tool to Disseminate Shared Savings Contracts

We explored the process of physician selection, focusing on selection of surgeons for knee and hip replacement to increase the probability of a new relationship, making cost and quality scorecard information more relevant.

 We collected data using a mailed survey sent to patients with knee or hip replacement surgery shortly after March 1, 2010. This time period followed a period of publicity about the new cost and quality scorecard.

 We used multivariate probit models to predict awareness of the scorecard and willingness to switch providers. Multinomial logit methods were used to predict the primary factor influencing the choice of surgeon (physician referral, family or friend referral, surgeon location, previous experience with the surgeon, or other).

Internet access and higher neighborhood incomes are associated with an increased probability of being aware of the scorecards. Male patients and patients with Internet access or in highly educated neighborhoods are more likely to be willing to switch providers for a reduced copay. Urban residents are more likely to rely on physician referrals, and rural patients on family/friend referrals when selecting a surgeon; Internet access reduces importance of surgeon location.

 Additional research is needed to determine whether Internet access is causal in improved responsiveness to market information and incentives, or a proxy for other factors. In addition, we see evidence that efforts to improve healthcare quality and costs through market forces should be tailored to the patient’s place of residence.

		Two patient characteristics rarely incorporated in this literature, Internet access and urban/rural residence, are strongly related to our dependent variables: awarenessof the scorecard, willingness to switch providers, and factors influencing choice of surgeon.

		Internet access is associated with greater responsiveness to market forces. Additional research is needed to determine whether this is causal, or a proxy for unobservedvariables.

		Information used to select physicians varies by patient’s rural/urban status, suggesting that efforts to use market forces to drive improvements in cost and quality should be specific to the patient’s geography.

The increased focus on use of market forces to contain costs and improve quality of healthcare has raised questions about the ability and willingness of consumers to make informed choices in this complex area. Significant variance in the cost and quality of care provided by physicians has been well documented,

 offering the potential for real choice on the part of patient-consumers and scope for the usefulness of cost and quality information. In an effort to bring these market forces to bear, policy makers have been encouraging the provision of cost and quality information, and the use of increased patient cost sharing to heighten awareness of cost differences.

For quality information to influence physician choice, (1) the patient must be aware of and have access to the information, (2) they must trust the information source, (3) it must be understandable, and (4) the information must be salient to the choice. Because of measurement issues, quality information has historically been focused at the health plan and hospital level, but physician-level information is becoming increasingly available,

typically through the Internet. Despite its increasing prevalence, consumer awareness of physician quality information remain slow; the most recent update of the Kaiser Family Foundation survey on the topic

 found that only 12% of consumers were aware of a source of physician quality information. Others have found that employer communication efforts are effective in increasing awareness of provider quality information in employment-based insurance.

Consumer trust of the information and the sponsors is an important issue. Patients consistently express strong trust in recommendations from personal sources, such as family or friends,

 with formal sources of quality information and professional recommendations ranked below personal sources. Of the formal sources of quality information described, Alexander et al

 found that 37% of consumers “trust information a lot” from a health insurance company, with an additional 44% expressing some trust in insurance companies as a source of quality information, expressing significantly more trust for  insurers than for employers or government agencies. Because health insurers have access to both contractual costs and the ability to evaluate processbased measures of quality, they are poised to leverage this trust in their covered population.

There is a rich literature examining the impact of format on the use of quality information and the consumer’s ability to comprehend the information presented. Simmeasures such as star rankings are preferred.

Providing a definition of quality and context for the quality information have also been found to be important,

 improving saliency and the value consumers place on the information. Additionally, providing cost and quality information simultaneously helps the consumer avoid interpreting higher costs as a signal for quality,

 improving the consumer’s ability to accurately identify high-value providers.

The interplay of cost incentives and quality information is key to the use of market forces to improve value in healthcare delivery. Harris

 found that privately insured respondents would accept a more restrictive provider network in exchange for higher quality of care, but suggests that the level of quality required for patients to accept some of the restrictions is probably unattainable. For example, disallowing self-referral to specialists required huge percentage point improvements in quality measures to compensate for the change, such as the fraction of patients extremely satisfied with their involvement in care decisions (+57%) or fraction of physicians rated excellent by medical experts (+52%). More recently, a stated choice experiment by Sinaiko

 looked at the combined impact of cost and quality information. This work suggests that copay differences can induce a modest amount of movement to a preferred physician tier, unless the patient has received a personal or professional referral for a physician in a nonpreferred tier.

Our study was designed to build on this previous work by analyzing the influence of a health plan—sponsored physician rating program on physician selection. In order to maximize the saliency of cost and quality information, we focused our attention on patients who were searching for an orthopedic surgeon for hip or knee replacement surgery in the months after this new Internet-based information source was publicized by both the health plan and the local newspaper. The information combines cost and quality information in a simple star rating, minimizing cognitive demands. This jointreplacement population was of interest, as these are often new patient–physician relationships and the procedure is rarely emergent, thus patients were more likely to search for and be influenced by cost and quality information.

Our study evaluated the use of a physician rating program developed by a health plan that provided physician cost and quality of care measures to their enrollees during 2010. Physicians received 1 star if they were above the national mean for quality, and a second star if they were also below the local benchmark for costs. Of the 9472 doctors rated, 20% received 1 star and 49% received 2 stars. The following is an excerpt of the health plan’s description of the program: “Evaluation for quality compares a physician’s observed practice to a national rate among other physicians who are responsible for the same interventions. Evaluation for cost efficiency compares a physician’s observed episode costs to the risk-adjusted costs of their peers in the same specialty and market… Physicians must first be designated for quality in order to be designated for cost efficiency… Quality and cost-efficiency evaluations both incorporate adjustments for case mix and severity of illness where appropriate.”

Our sample included patients who had knee or hip replacement surgery shortly after the physician rating program was publicized. The first 400 patients from each category (hip/knee) were selected by date of service starting March 1, 2010. All of these enrollees received their services during 2010 three months after the rating program was initiated. The mailed surveys included a $2 bill to encourage participation; no follow-up with non-respondents was possible because the responses were anonymous. This project was reviewed and approved by the Western Institutional Review Board.

The survey questionnaire included the following questions:

		Are you aware of the physician scorecard ratings on cost and quality performance provided by your health insurance plan?

		If so did you use this information when selecting the surgeon for your knee/hip surgery? (This was specific to each respondent's group.)

		If you did not use the physician rating information, how did you select your physician? (Family/friend information, physician referral, physician location, network restrictions, physician gender, language spoken by physician, or other; open response for specification of “other.”)

		If you were offered a copay reduction to switch to a lower cost physician of equal or higher quality would you do so?

		Do you have access to the Internet at home?

		Do you have access to the Internet at work?

Response rates were adequate at 60%, and were fairly consistent across demographic groups, ranging from 55% to 67% when subdivided by age, sex, urban status, and procedure type. The only demographic group that dropped below a 55% response rate was the Medicaid population; the 36% response rate dropped the Medicaid population from 9% to 5% of the respondents. This lower response is not surprising given the more transient residential status of a Medicaid population.

From the health claims data, the health plan’s informatics team collected the following information to merge with the survey results: age, sex, procedure type (hip/knee), urban/rural indicator, type of insurance (commercial, Medicare, Medicaid, or Medicare/Medicaid dual eligible). In addition, the informatics team included the median education and income level in the respondent’s census tract, based on the 2010 US Census, stripping identifiers before delivering the data set to the research team. To preserve the deidentified status of the data, median income in the census tract was rounded to the nearest $500.

Median education was measured using the percentage distribution of residents in the census tract among 3 categories: no high school degree; high school degree but less than 4-year college degree; and 4-year college degree or higher. The extreme categories, no high school and 4-year degree or higher,were highly skewed. We parameterized census tract education as a trinomial variable to capture whether a census tract was in the tail of the skewed variables or not, selecting the following 3 categories: (1) high average education—more than 50% of the residents in the census tract have a 4-year degree; (2) low average education&mdash;more than 10% of the residents in the census tract did not achieve a high school degree; and (3) baseline average education&mdash;at least 10% have a high school degree and fewer than 50% have a 4-year degree.

Summary statistics for the independent and dependent variables are shown in 

. There is a tendency for the average respondent to be female (64%), live in an urban area (66%), and have access to the Internet (69%). With an average age of 68 years (range 23-93), the bulk of the respondents (64%) are covered under a Medicare plan, which is to be expected given the nature of the surgeries. The majority of respondents live in census tracts that are in the baseline education category, with 28% in low-education census tracts and 19% in high-education census tracts. To capture income, we generated a categorical variable that indicated the quartiles of the census tract median income, with break points rounded to $47,000, $57,000, and $71,000.

Awareness of the new scorecard was still quite low, at 11%, and a small minority of these (only 7 respondents) said they used the scorecard information in the selection of their surgeon. There does appear to be a significant fraction of the population that has some price sensitivity, with 30% of the respondents indicating some willingness to switch physicians for a lower copayment, assuming quality doesn’t decline.

Although the survey asked the respondent to indicate a single factor influencing their choice of surgeon, a fraction (22%) of respondents checked more than 1 factor. When there were multiple responses, we randomly selected 1 response for our analysis. Very few respondents indicated that physician gender, physician language, or physician network status was important in their selection process, so we merged these respondents into the “other” category. Despite our efforts to find participants who were making a new physician selection, 10% of the respondents indicated in the open response that they had had a previous experience with the surgeon. Therefore, we identified this as a new category, resulting in these final response categories: family/friend information (29%), physician referral (37%), physician location (13%), previous experience (10%), and other (11%).

Because so few actually used the scorecard, we did not investigate what characteristics were associated with scorecard use. Therefore, we focused on awareness of the scorecard, willingness to switch physician for a financial incentive, and the  selection process. The first 2 questions are dichotomous yes/no answers, modeled as a probit equation. We used multinomial logit to model the physician choice process. Both probit and multinomial logit estimations were completed using Stata 12. We explored nesting in physician choice, but found no improvement in fit or evidence of nesting in the choice process. We also tested methods15 of accounting for potentially endogenous variables (eg, access to the Internet and awareness of the scorecard), but found no evidence of bias due to endogeneity.

When modeling awareness of the physician scorecard (

), we found that the participant characteristic that has the greatest statistical significance is access to the Internet. Holding all other characteristics at their true values, moving the population from “no Internet access” to “Internet access” nearly doubled the average predicted probability of awareness of the scorecard, leading to the +6.5 percentagepoint (pp) marginal effect (ME) shown in the second column of Table 2. We also identified marginal statistical significance in the census tract’s median income. When we computed the marginal effect of moving the population from the lowest to highest median income quartile, we found a +9.1 pp ME, more than doubling the probability of being aware of the scorecard.

) appears to be sensitive to respondent’s sex, their access to the Internet, and the education level in their census tract. Females were much less likely to be willing to switch; the average predicted probability of being willing to switch drops by a —9.9 pp ME when moving from a male to a female status. Those who live in a highly educated census tract were much less willing to switch; moving from a baseline education to a highly educated census tract caused the willingness to change surgeons to drop by a –14.4 pp ME. In contrast, those with access to the Internet were more flexible in provider choice. Moving from “no access” to “Internet access” generated a +8.5 pp ME.

Two of our respondent characteristics appear to have significantly impacted the physician selection process (

): rural/urban residence and access to the Internet both have a χ2 P value less than .05. The marginal effects in 

 show us that those who live in an urban area were much less likely to rely on referrals from families and friends (—11.2 pp ME), and much more likely to rely on physician referrals (+15.6 pp ME). Those with access to the Internet were much less likely to rely on the physician’s location in their selection (–12.5 pp ME), with an increased reliance on previous physician experience (+5.4 pp ME) and other factors (+5.5 pp ME) to influence their choice.

It is interesting that we found no predictive value of age or type of insurance product (commercial, Medicare, Medicaid) when modeling the 3 dependent variables—awareness of the scorecard, willingness to switch providers for a lower copay without sacrificing quality, and the factors influencing choice of surgeon. On the other hand, Internet access, a variable rarely included in this literature, was strongly associated with all 3 dependent variables. It is possible that this is a causal factor&mdash;the greater informational resources that come with access to the Internet may alter the choices and attitudes of the respondent. With Internet access, the consumer can more easily find cost and quality information,

as well as search for other, non-clinical quality dimensions such as practice size, presence of urgent care facilities, and hospital affiliation. This suggests that improving access to the Internet, or reducing search costs in other ways, such as employer-facilitated provision of quality information,

may improve awareness and use of quality information, leading to increased patient-driven market forces. Alternatively, Internet access may be a proxy for (latent) respondent characteristics such as generalized differences in the way they engage with their environment. This is an area that deserves additional investigation. If it is a causal factor, this has great implications for the impact of market forces as Internet access increases in prevalence, particularly through mobile devices.

This study is unusual in its inclusion of rural versus urban status, which we find is an important factor in the source of informationused by the respondent in physician selection. The greater importance of family and friend referrals in rural locations implies stronger reliance on social fabric in rural areas, while the urban respondents had a much greater tendency touse professional referrals in their choice of surgeon. This suggests that, as we seek to increase the impact of market forces to drive improvements in cost and quality, different strategies may be needed by geography.

By selecting first observances of a claim for hip and knee surgery, this project was designed to highlight instances of orthopedic surgeon selection when an existing relationship is unlikely. This was intended to maximize the probability that quality and cost data provided by the health plan would be salient, in a setting where the patient had time for research, physician cost and quality rating information had been recently publicized, and the data were easily accessible on the Internet. Yet only 11% of the respondents indicated that they knew about the physician rating program, consistent with surveys of awareness in a more general population.

 Since nearly 70% of the respondents indicated that they have access to the Internet at home or office, the media for transmission of the cost and quality information does not appear to be the primary limiting factor. Given these conditions, we must conclude that either the health plan enrollees in this study do not place a high value on cost and quality data when selecting their surgeon, or that the message regarding the availability of the information was ineffective.

The availability of “product quality” information is a key feature of a functioning market. Therefore, if healthcare cost and quality improvements are to be driven by market forces, patient awareness and use of physician quality information are necessary. This work identifies factors that are important in patient awareness of quality information, and explores the physician selection process more broadly. Our results suggest that easy access to information through the Internet is a key component of awareness of quality information and willingness to change providers, and reduces the patient’s reliance on proximity to physician. This implies that other efforts that reduce search cost may be important, such as employer provision of information orstate-based public and private initiatives, such as those tied to the Robert Wood Johnson Foundation’s Aligning Forces for Quality initiative. We also find that cultural differences, here captured in rural versus urban residence, should influence communication strategies as they influence which factors are important to decision makers.

Because we focused on orthopedic surgery, the population was fairly old (average age 68.6 years), so these results may not be generalizable to a younger population. Age and type of coverage (Medicare vs commercial insurance) were not found to be statistically significant in any model, but there may be a subtle impact not detected. To capture a more age-diverse population, we originally attempted to survey new mothers about their choice of pediatrician, because these were also new relationships formed in a non-emergent setting, but—perhaps due to the demands of new motherhood&mdash;the response rate for this population was abnormally low (under 10%) and the data for new mothers had to be excluded. 

 From Medica Research Institute (CSC, JK, MS), Minneapolis, MN; University of Minnesota (CSC, JK), Minneapolis, MN.

This project was funded by an internal award from the Medica Research Institute.

 The authors (CSC, JK, MS) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Concept and design (JK, MS); acquisition of data (CSC, JK, MS); analysis and interpretation of data (CSC, JK, MS); drafting of the manuscript (CSC, JK, MS); critical revision of the manuscript for important intellectual content (CSC, MS); statistical analysis (CSC, MS); obtaining funding (JK); administrative, technical, or logistic support (MS); and supervision (CSC, JK).

Caroline S. Carlin, PhD, Research Investigator, Medica Research Institute, MR-CW295, PO Box 9310, Minneapolis, MN 55440. E-mail: caroline.carlin@medica.com.1. Kralewski JE, Dowd BE, Xu Y, Knutson D. The organizational characteristics of best medical group practices. 2011. A Robert Wood Johnson HCFO project, Grant Number 065450.

This article explores the information (scorecards, referrals, prior experience, location) influencing the choice of orthopedic surgeon and how this influence varies by patient characteristics.

Sources of Information Used in Selection of Surgeons

 After a new formulation of levothyroxine was distributed in Israel, side effects were reported to the Ministry of Health generating extensive media coverage. The purpose of this study was to determine whether the new formulation was associated with a change in thyroid-stimulating hormone (TSH) levels of treated patients and to evaluate the effect of the extensive media coverage on the incidence of laboratory test performance.

Retrospective-cohort and crosssectional analysis.

All patients from the Leumit Health Services of Israel treated with levothyroxine between October 2009 and February 2012 were included in the study. A retrospective cohort was constructed of patients treated and maintained within the desired target range (0.35-5 mIU/L) from January to July 2010. A longitudinal analysis was conducted to calculate the monthly distribution of TSH levels from laboratory tests during routine care over 26 months. Data were stratified by cohort and noncohort patients.

 Data were captured for 18,106 levothyroxine-treated patients; 1140 were included into the retrospective cohort. In both subpopulations a sharp rise in the number of tests performed monthly is observed at the peak of media coverage during  October and November 2011. In the retrospective cohort the proportion of TSH results within target range fell to a low of 67.5% during December 2011, with 25.3% between 5.01 and 20 mIU/L. Results >20 mIU/L then peaked at 3.8% indicating an increase in patients who stopped taking levothyroxine.

 These results demonstrate the power of mass media to influence patient behavior and to foment a public health  scare.

Stress induced by inaccurate or misunderstood messages from media coverage may have caused patients to stop taking an important drug necessary for their health and well-being.

		Medical decision makers should not underestimate the power of the mass media to both influence patient behavior and to foment a public health scare.

		Methods must be developed by providers to convey accurate information to both patients and healthcare providers when questions arise concerning the safety and efficacy of widely used products.

		Managed care information systems can provide the population-based data necessary to scientifically evaluate changes in patient response to new products.

Primary hypothyroidism is the most common endocrine disease.

The issue of bioequivalence studies for narrow therapeutic index drugs, such as levothyroxine, has been a subject of debate.

Adverse outcomes have been associated with the approved generic substitution of levothyroxine products, frequently without the prescribing physician’s knowledge.

About 250,000 Israelis are treated for hypothyroidism with the drug levothyroxine, almost exclusively with the brand product Eltroxin, manufactured by GlaxoSmithKline. During the month of February 2011, a new formulation of Eltroxin was distributed in Israel in which changes had been made in the tablets’ inert ingredients. Previously, an eruption in adverse effects reports was observed in New Zealand after a new formulation of Eltroxin had been introduced in 2007.

However, this experience was not reported to the Pharmacy Division of the Israel Ministry of Health when the new product was registered in Israel in 2008 since  the manufacturer dismissed the incident as being provoked by the media and less associated with the change in formulation. In September 2009, when GlaxoSmithKline sold the production of Eltroxin to Aspen Pharma, Perrigo Israel acquired the registration rights for the product in Israel. Upon receiving Ministry of Health approval for the new formulation in February 2011, the new product was made available to pharmacies in Israel. Shortly after the new formulation was marketed, cases of side  effects began to be reported to the Ministry of Health which included fatigue, weakness, depression, tremors, diarrhea, and arrhythmias. These reports generated extensive media coverage that reached peak intensity during the month of October 2011. Consequently, growing alarm and confusion emerged among treated patients and their families, resulting in a public outcry  accusing all stakeholders involved of negligence and malfeasance.

In response, the director general of the Ministry of Health commissioned a board of inquiry. Although the commission was composed of both clinicians and regulators, the scope of the investigation was mostly limited to the registration process of the product and to the sporadic reports of side effects and complications.

 Of the approximately 800 reports received, only 105 (13.2%) mentioned thyroid-stimulating hormone (TSH) levels, 85 reported TSH elevation, 6 reported a drop, and 14 reported only a “change.” Laboratory data corroborating the report were received in only 35 of the reports, rendering large, population-based analyses unfeasible. To the best of our knowledge, the results of a longitudinal obrelevant period have not been reported.

We hypothesized that if the bioavailability of the new product was significantly inferior to that of its predecessor, measurable populationwide variance in TSH test results would be observed among levothyroxine-treated patients before and after the introduction of the new formula. Additionally, we postulated that the intense media coverage may have affected patient behavior both by inducing compliant patients to perform additional TSH tests and by motivating negligent patients to request blood tests at health maintenance organization (HMO) clinics. We also suspected that some patients, confused by the abundance of conflicting information, may have stopped taking Eltroxin, fearing that the new product was indeed unsafe. The purpose of this study was to determine whether the change in the formulation and later, the intense media coverage, were associated with a measurable change in TSH levels of treated patients, and to evaluate the effect of the extensive media coverage on the incidence of laboratory test performance.

This study was conducted in the Leumit Health Services (LHS), a managed care organization that provides medical care and coverage to approximately 700,000 members throughout Israel. Since 1999, LHS has implemented an electronic patient record (EPR) system. During each patient visit, all physicians complete an EPR detailing the specific clinical services provided including diagnoses, drugs prescribed, and diagnostic tests ordered. Additionally, LHS operates a central laboratory which transmits all test results to a central data warehouse that interfaces with the EPR program. Similarly, all data on prescriptions dispensed in LHS-operated community pharmacies and in private pharmacies contracting with LHS are also transmitted to the data warehouse.

To test our research hypotheses and answer our primary research questions we implemented 3 individual study designs, as follows:

1. Population-based, cross-sectional analysis of TSH levels and frequency of testing in levothyroxine-treated patients.

To evaluate the number of TSH tests performed monthly by all patients in the HMO and the distribution of test results throughout the study period, we implemented a cross-sectional study. All patients treated with levothyroxine between October 1, 2009, and February 29, 2012, were included. Since Leumit patients can purchase a 3-month supply of levothyroxine, prescribing data from October to December 2009 were included to identify all patients treated during January 2010. The number of tests performed, mean TSH results with 95% confidence intervals, and median TSH results were calculated for monthly results recorded over the study period (January 2010 until February 2012).

2. Retrospective cohort study of TSH levels and frequency of testing in levothyroxine-treated patients.

To eliminate incidence/prevalence bias, a retrospective cohort was constructed of patients with a record of at least 1 dispensed prescription for levothyroxine during the 4 months prior to February 1, 2010, and a recorded TSH laboratory test result within the target range (0.35-5 mIU/L) during January 2010. Furthermore, to avoid biases emanating from cases difficult to titrate or nonadherent patients, subjects were excluded from the cohort if additional test results not within target range were recorded between January 1, 2010, and June 30, 2010. A longitudinal analysis was conducted to calculate the monthly distribution of TSH levels as recorded from laboratory tests performed during routine care as previously described for the population-based cross-sectional analysis. Additionally, TSH levels were recoded into 4 subcategories (

0.34, 0.35-5.0, 5.1-20, and >20 mIU/mL). The monthly distributions of test results across the 4 categories in the retrospective cohort population were calculated and statistical significance was tested with the 2-sided χ

 value <.05 was considered statistically significant.

3. An evaluation of prevalence of TSH testing among untested levothyroxine-treated patients after the intense media coverage.

To determine whether intense media coverage motivated previously nonadherent patients (ie, nonadherence to laboratory testing) to perform TSH testing, we identified patients continuously treated with levothyroxine throughout the 12 months prior to the peak in media coverage (October 2011), but had no record of a TSH test recorded in their EPR during this 12-month period. Patients were included if dispensed prescriptions for levothyroxine were recorded continuously from June 2010 to November 2011, but a TSH test was not recorded in the EPR during the 12 months prior to October 2011 when the intense media coverage appeared. We calculated the number and proportion of these patients who performed at least 1 test during October and November 2011.

Drug dispensing and laboratory test data were captured from “LHS’s” data warehouse using IBM Cognos 8 BI Report Studio software. Results of queries were downloaded into Microsoft Excel spreadsheets and data analysis was performed using IBM SPSS Statistics version 20 software. A chart review was performed on a sample of electronic records of patients in the target population to glean relevant information recorded by physicians in free text. Ethics committee approval was obtained from LHS’s institutional review board.

Laboratory data were captured for 18,106 levothyroxine-treated patients during the study period, 1140 of whom met the criteria for inclusion into the retrospective cohort (

). Over the 26-month period 75,082 and 6328 tests were performed for the patients in the noncohort and cohort populations, respectively. The number of TSH tests performed monthly for each subpopulation is shown in 

. In both subpopulations a sharp rise in the number of tests performed monthly is observed at the peak of the media coverage during the months of October and November 2011, dropping during the following month. The mean and median TSH results for both subpopulations per month appear in 

, respectively. A gradual nonstatistically significant increase in mean TSH values is observed in both populations after February 2011 with a steeper rise observed after October, peaking during December, and then decreasing during January 2012. This trend is particularly pronounced in the cohort population where mean TSH was 3.93 mIU/L during October 2011, 5.44 mIU/L in December, then falling to 3.77 mIU/L in January 2012. Median TSH was calculated to be 2.81 mIU/L during October 2011, peaking at 3.57 mIU/L during December, and then falling to 2.91 in January 2012. This trend is further illustrated by the TSH value measured for the third quartile which only surpasses the 5 mIU/L mark (the upper boundary of the desired target range) during the 2 months following the intense media coverage, and then falling to 4.51 mIU/L during January 2012. The monthly distribution of TSH test stratified by range for the retrospective cohort is presented in 

During February 2011, the month the new product was introduced, 84.8% of the tests were within the desired target range, 13.8% between 5.01 and 20 mIU/L, and 0.5% >20 mIU/L. The proportion of test results within target range fell to a low of 67.5% during December 2011, with 25.3% between 5.01 and 20 mIU/L and the proportion of test results >20 mIU/L peaking at 3.8%. The distribution observed 1 month later during January 2012 was similar to that observed the previous September before the story hit the press.

We identified 1078 patients treated with levothyroxine for at least 12 months prior to October 2011 for whom there is no record of a TSH test result in their EPR during this period. Of these patients, 462 (42.9%) performed at least 1 TSH test during the months of October and November 2011.

To the best of our knowledge this study is the first population-based quantitative analysis of patient response to a mass media—induced panic concerning a prescription drug for treatment of a chronic disease. Although a gradual rise in TSH levels is observed over time, clinically significant changes in mean, median, and third-quartile TSH levels are first observed the month after the story was extensively publicized. While not indicating that the new formulation brought about prevalent, significant changes in patient health statuses, these findings do indicate that the behavior of portions of patients was significantly affected by the abundance of conflicting information to which they were exposed. It is noteworthy that the changes observed in mean TSH values observed from October to December resulted predominantly from outlier values >20 mIU/L, a result which generally indicates that the patient has stopped taking levothyroxine. (We performed a chart review of a small sample of these patients and in almost all cases the physician had recorded in free text that the patient had ceased taking medication at their own decision.) This finding is substantiated by the prevalence of test results >20 mIU/L which peaked to 3.5% in the cohort population during December 2011, the same month that the mean and median reached their peaks. Accordingly, the significant changes observed between October and December 2011 can be attributed predominantly to patients who, confused by the deluge of information, stopped taking their medication. We interpret these findings as such since, had these patients stopped taking levothyroxine without connection to the media reports, we would have seen similar behavior throughout the study period and not distinctively at this point in time.

This premise is supported by the trends observed in prevalence of laboratory test performance over the 26-month study period. Presumably, had the bioavailability of the new product been significantly inferior to its predecessor, we would have expected to observe an increase in the number of tests performed among patients seeking medical care and consultation upon the advent of signs and symptoms of hypothyroidisms approximately 6 weeks after the new product had been marketed. However, such behavior was not observed in this study population, either in the cohort or in the general population. Over the 26-month study period substantial increases in prevalence of TSH testing is first observed during October 2011, peaking during November and then subsiding during December. These results should however be evaluated with caution since they are confounded by the occurrence of the Jewish high holidays which in 2011 fell mostly in October. In Israel these are national holidays when HMO clinics do not operate, all government offices and most businesses are closed, and Israelis are busy with preparations for the holidays or travel abroad. The crude data for this month therefore do not accurately reflect the magnitude of the phenomenon. When the number of tests performed by the cohort population in October 2011(N = 317) is compared with those for September 2010, the month of the holidays that year (N = 123), a 258% increase in testing prevalence is calculated. This steep increase in utilization of laboratory resources during the holiday season vividly reflects the level of concern prevalent at that time. Accordingly, the results of this nationwide study do not support the allegation that the new product was seriously inferior to its predecessor. These results do, however, demonstrate the power of the mass media to both influence patient behavior and to foment a public health scare. The large proportion of patients untested for at least 1 year who sought care when the issue became a matter of major public concern provides lucid testimony of this phenomenon.

Our study has several strengths absent in the report from New Zealand.7 First, the study data were captured from a data warehouse of an HMO which has fully implemented an EPR system among all practicing physicians. Since this system interfaces with the HMO’s laboratory and pharmacies we had all prescription and laboratory data necessary to evaluate changes in TSH levels over time in a defined patient population. Second, these data capabilities facilitated the implementationof a robust study design which included both a cross-sectional analysis and a retrospective cohort design. By including both designs in this study we were able to conduct a longitudinal analysis of testing prevalence among all treated patients while simultaneously presenting results for a subpopulation without biases caused by incident or difficult to titrate cases. Third, since LHF provides services to patients nationwide, this study included patients treated throughout Israel from all major ethnic and socioeconomic subpopulations. Limitations of this study include lack of laboratory data from hospitalized patients and deficiencies in surveillance due to patient nonadherence to TSH test ordering. Furthermore, since the new formulation had the same product code as the previous product we could not determine when patients started to take the new product. Similarly, the results obtained in the first weeks after the new product was introduced may have been biased by patients and pharmacies that had not yet finished supplies of the old formulation.

The major lesson learned from this analysis is that stress induced by inaccurate or misunderstood messages from media coverage may cause patients to stop taking an important drug necessary for their health and well-being. Since we have no expectations that media coverage in the future will be more responsible, methods must be developed by providers to convey accurate information to both patients and healthcare providers when questions arise concerning the safety and efficacy of widely used products. Additionally, this study demonstrates the potential contribution of healthcare information systems to provide the population-based data necessary to evaluate changes in patient response to new products. With generic switching and introduction of new formulations becoming more common internationally, the methodologies implemented in this study may be useful in future situations in other countries when changes in formulations or generic switching initiate a public scare.

 From Medical Division, Leumit Health Services (NRK, DAV, DAW, GB), Tel-Aviv, Israel; School of Public Health (NRK), Sackler Faculty of Medicine, Tel-Aviv University, Tel-Aviv, Israel; Faculty of Health Sciences (DAV, GB), Ben-Gurion University of the Negev, Beer-Sheva, Israel.

 The authors (NRK, DAV, DAW, GB) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (NRK, GB); acquisition of data (NRK); analysis and interpretation of data (NRK, DAW, GB); drafting of the manuscript (NRK); critical revision of the manuscript for important intellectual content (NRK, DAV, DAW, GB); statistical analysis (NRK); administrative, technical, or logistic support (DAV, DAW); and supervision (DAV).

Natan R. Kahan, PhD, RPh, MHA, Medical Division, Leumit Health Services, 23 Schprintsak St, Tel-Aviv, 64738, Israel. E-mail: nkahan@post.tau.ac.il.1. Chakera AJ, Pearce SH, Vaidya B. Treatment for primary hypothyroidism: current approaches and future possibilities. Drug Des Devel Ther. 2012;6:1-11.

2. Blakesley VA. Current methodology to assess bioequivalence of levothyroxine sodium products is inadequate. AAPS J. 2005;7(1): E42-E46.

3. Bolton S. Bioequivalence studies for levothyroxine. AAPS J. 2005; 7(1):E47-E53.

4. Henderson JD, Esham RH. Generic substitution: issues for problematic drugs. South Med J. 2001;94(12):1235.

5. Tanne JH. US doctors and patients are split on approval of generic thyroid hormone. BMJ. 2004 ;329(7459):192.

6. Hennessey JV, Malabanan AO, Haugen BR, Levy EG. Adverse event reporting in patients treated with levothyroxine: results of the pharmacovigilance task force survey of the american thyroid association, american association of clinical endocrinologists, and the endocrine society. Endocr Pract. 2010;16:357-370.

7. Faasse, K, Cundy T, Petrie K. Medicine and the media: thyroxine: anatomy of a health scare. BMJ. 2009;339:b5613.

8.Linder-Ganz R, Even D. Israel Health Ministry to investigate handling of drug’s side effects. Haaretz. http://www.haaretz.com/print-edition/news/israel-health-ministry-to-investigate-handling-of-drug-s-sideeffects-1.392354. Published October 28, 2011. Accessed April 16, 2012.

9. Linder-Ganz R, Even D. Israel drug company summoned to hearing over thyroid drug Eltroxin. Haaretz. http://www.haaretz.com/news/national/israel-drug-company-summoned-to-hearing-over-thyroid-drugeltroxin-1.409067. Published January 25, 2012. Accessed April 16, 2012.

10. A change in the formulation of Eltroxin: Report of the Commission of Inquiry. Presented to Prof. Roni Gamzu, Director General, Ministryof Health. January 8, 2012.

Sporadic, unsubstantiated side effects were reported in the mass media after a formulation change of levothyroxine tablets induced patients to monitor thyroid-stimulating hormone levels and to unnecessarily stop taking medication.

A Natural Experiment in Mass Media Modulated Pharmacokinetics After a Change in Tablet Formulation

Limitations of current diagnotic techniques may allow some patients with presumed renal cell carcinoma (RCC) to undergo nephrectomy without definitive confirmation of malignancy.

 To confirm previous estimates of postnephrectomy renal mass diagnosis and to assess the economic impact of nephrectomy.

 This retrospective cohort analysis identified commercial enrollees who underwent nephrectomy with a diagnosis of RCC between July 1, 2000, and March 30, 2008. Study subjects were stratified based on medical claims for benign or malignant disease after the nephrectomy date. Cohorts were compared on resource utilization before and after nephrectomy, occurrence of postsurgical complications, and associated 1-year costs of care.

Of 10,404 patients undergoing nephrectomy for presumed RCC, 1613 (15.5%) were subsequently identified as having benign disease, despite median presurgical diagnostic expenditures of $1311 per patient (interquartile range [IQR], $467-$2606). Median expenditures for the 12 months postnephrectomy were $26,920 per patient (IQR, $16,851-$46,982) for those with malignant disease  and $23,951 per patient (IQR, $14,873-$38,190) for those with benign disease (

<.0001). For patients with benign disease, 17.5% experienced a postsurgical adverse event, resulting in a 1.5-fold increase in expenditures (median $31,838 per patient for those with event vs $22,770 per patient for those without event; 

 In this study, approximately 1 in 6 patients were found to have a benign renal mass  postnephrectomy. Given the risk of surgical complications and related economic consequences, methods for better identifying malignant versus benign disease prior to surgery could provide significant benefits to patients and payers.

This analysis of insured Americans suggests that 1 in 6 patients who undergo nephrectomy for suspected renal cell carcinoma are subsequently found to have benign disease.

		The median cost of this care, inclusive of and within 12 months of surgery, was $26,405 per patient.

		Among those with complications after nephrectomy, expenditures may exceed $40,000 per patient in the year following surgery.

		Assuming 58,000 new renal tumors will be detected annually using currently available diagnostic techniques in the United States and treated surgically without tissue confirmation of cancer, the American healthcare system might be resourcing upward of 10,000 potentially avoidable nephrectomies each year.

Renal cell carcinoma (RCC) accounts for approximately 3% of adult malignancies and more than 80% of kidney cancers, causing approximately 13,000 deaths in the United States each year.

 It is typically characterized by a lack of early warning signs, diverse clinical manifestations, and tumor resistance to radiation and traditional chemotherapy.

The increasingly prevalent use of cross-sectional imaging for nonspecific abdominal complaints has resulted in the earlier diagnosis of RCC and a preponderance of small asymptomatic lesions. Thus, the continual rise in RCC incidence is attributed largely to enhanced detection, with more than 70% of kidney cancers now being discovered incidentally and at a size less than 7 cm.

 Nephron-sparing surgery is gaining acceptance as the treatment of choice for these smaller tumors, while radical nephrectomy remains the standard for larger tumors not amenable to partial nephrectomy.

Differentiating between benign and malignant renal tumors through current radiographic or clinical methods remains a challenge. While renal biopsy for molecular analysis of tissues might provide additional information in this setting, it yields nondiagnostic information in up to 20% of cases and is often available only at academic centers.

Importantly, previous research suggests that approximately 20% of patients who undergo nephrectomy are found to have benign postoperative tumor histology, implying the possibility of surgical overtreatment.

Based on current estimates of 58,000 patients with new RCC diagnoses in the United States each year, the vast majority of whom will undergo nephrectomy prior to a definitive tissue confirmation of cancer, it is conceivable that the American healthcare system is resourcing upward of 10,000 potentially avoidable nephrectomies each year.

The objectives of this analysis were to confirm previous estimates of postnephrectomy renal mass diagnosis and to assess the economic impact of nephrectomy, both in terms of surgical costs and resource utilization related to postoperative complications, in a large population of commercially insured Americans.

Data were obtained from the IMS LifeLink Database, which at the time of the study was composed of medical and pharmaceutical claims for approximately 60 million unique patients from 100 health plans, including health maintenance organizations (HMOs), preferred provider organizations (PPOs), point-of-service (POS) plans, indemnity plans, and other types, located throughout the United States. The database includes both inpatient and outpatient diagnoses (

International Classification of Diseases, Ninth Revision, Clinical Modification

] codes) and procedures (Current Procedural Terminology, Version 4 [CPT-4] and Healthcare Common Procedure Coding System codes), as well as both standard and mail order pharmacy records. Data elements include demographic variables (age, sex, geographic region), health plan type (eg, HMO, PPO, POS), payer type (eg, commercial, self-pay), provider specialty, reimbursed amount for service, date of service, and start and stop dates for plan enrollment.

Because all pertinent patient information in the database is encrypted and de-identified, and no patient contact was involved, no informed consent or approval by an institutional review board was required or sought. The data source was fully Health Insurance Portability and Accountability compliant.

Patients were eligible for analysis if they had at least 2 medical claims for RCC (defined as 

codes 189.0, 189.1, 198.0, 236.91) during the July 1, 2000, through March 30, 2008, database extraction period. Patients were further required to have a claim for nephrectomy (CPT-4 codes 50220, 50225, 50230, 50234, 50236, 50240, 50543, 50545, 50546, 50547, 50548) during the study period to be included in the analysis (note that the nephrectomy claim could coincide with 1 of the 2 required RCC diagnoses). A minimum of 2

 RCC diagnoses were required to ensure all patients included in the study presented with a high preoperative clinical suspicion of presumed renal malignancy.

The date of the nephrectomy claim during that period was assigned as the index date. From that date, 6 months of preindex and up to 12 months of postindex data were collected for all health plan—eligible patients. Patients had to have continuous health plan enrollment for 6 months before and at least 6 months after the index date to ensure complete baseline and follow-up information on all patients in the study.

Two mutually exclusive study groups were identified based on medical claims. Patients were categorized as ultimately having either benign disease or malignant disease following the nephrectomy. Benign disease was defined as any patient having a medical claim with a benign diagnosis (

 codes benign neoplasm, kidney, except pelvis [223.0], renal pelvis [223.1]) subsequent to the index date (nephrectomy).To be conservative regarding the prevalence of benign disease postnephrectomy, all other patients (ie, those with a minimum of 2 medical claims for a diagnosis of RCC, but with no benign diagnosis subsequent to the nephrectomy) were considered to have malignant disease.

The postnephrectomy benign and malignant cohorts were compared with respect to patient demographics (age, sex, region of the United States where surgery was performed), surgery type (nephron sparing vs radical nephrectomy), and comorbid conditions (using Charlson Comorbidity Index score, individual comorbidities contributing to the score).

 In addition, rates of nephrectomy-related adverse events were calculated for each cohort: hemorrhage (

 998.1x), deep vein thrombosis (ICD-9-CM 453.40, 453.41, 453.42), pulmonary embolism (

 415.11, 415.12, 415.19), myocardial infarction (

 410.x), surgical woundinfection or disruption (

Methods for better identifying malignant versus benign disease before nephrectomy could provide significant benefits to patients and payers.

August 2013

Option Pricing: A Flexible Tool to Disseminate Shared Savings Contracts

Sources of Information Used in Selection of Surgeons

A Natural Experiment in Mass Media Modulated Pharmacokinetics After a Change in Tablet Formulation

Cost of Care for Malignant and Benign Renal Masses

Long-Term Statin Use and the Risk of Parkinson's Disease

Potential Misuse and Inappropriate Prescription Practices Involving Opioid Analgesics

Does Medication Adherence Lead to Lower Healthcare Expenses for Patients With Diabetes?

False Activation of the Cardiac Catheterization Laboratory for Primary PCI