 Thyroid-stimulating immunoglobulins (TSIs) are autoantibodies that bind to the thyroid-stimulating hormone (TSH) receptor on thyroid cells, resulting in Graves’ disease (GD), the most common cause of hyperthyroidism. Recently published guidelines recognize the value of anti-TSH receptor antibodies, and a TSI test with high sensitivity and specificity for GD, recently cleared by the US Food and Drug Administration, is now available. Despite this, existing diagnostic algorithms for hyperthyroidism do not currently include TSI testing except in specific cases like pregnancy. The objectives of this analysis are to understand whether incorporating a test that specifically detects TSIs into existing algorithms results in cost savings and reduces time to diagnosis for payers and managed care organizations.

 An evidence-based economic model was developed to determine the average time to diagnosis and annual costs associated with various diagnostic algorithms for GD in a population of 100,000 managed care enrollees. Diagnostic algorithms used in current practice and hypothetical algorithms that include the TSI test were identified using published clinical guidelines and interviews with practicing endocrinologists. The model estimates costs of current and TSI test—based diagnostic algorithms using payment amounts for laboratory tests, procedures, and physician visits.

 Compared with non-TSI algorithms, 100% use of algorithms that include the TSI test result in 46% faster time to diagnosis and generate 47% overall cost savings due in large part to reductions in costly procedures and specialist office visits.

 Incorporation and early utilization of the TSI in vitro diagnostic test into current diagnostic algorithms confers cost savings and shortens time to diagnosis.

	Inclusion of a thyroid-stimulating immunoglobulin test earlier in the diagnostic pathway of patients with low thyroid-stimulating hormone indicative of hyperthyroidism results in significant cost and time savings to payers and at-risk managed care organizations.

		 Net direct costs of diagnosis are reduced by up to 43%.

		 Net cost of avoiding misdiagnosis and treatment of unexplained symptoms is reduced by up to 85%.

		 Overall net cost is reduced by up to 47% and time to diagnosis is reduced by as much as 46%.

Graves’ disease (GD) is one of the most common autoimmune diseases, affecting approximately 3.45 million people in the United States alone.

 In these individuals, thyroid-stimulating immunoglobulins (TSIs) are produced and directly contribute to GD pathology.

 It is estimated that up to 80% of hyperthyroid cases are caused by GD and, by inference, TSI; the remaining 20% of hyperthyroid cases are attributable to other non-autoimmune etiologies.

 Accurate diagnosis and treatment of GD early in the disease process improves patient quality of life by limiting many of the manifestations of GD including heart palpitations, irritability, anxiety, fatigue, and insomnia.

Differential diagnosis of hyperthyroidism, a systematic process of elimination to determine the underlying cause, can be an especially challenging and lengthy process. Many patients with hyperthyroidism present with atypical symptoms or have subclinical hyperthyroidism with no apparent symptoms, low to borderline-low thyroid-stimulating hormone (TSH), and normal free triiodothyronine (FT3) and free thyroxine (FT4) levels.

In these patients, diagnosis of GD may be delayed for months to years and this delay may adversely impact patient quality of life and productivity.

Not surprisingly, the diagnosis of patients with atypical symptoms usually centers on management of presenting symptoms; whereas, the underlying cause is neither diagnosed nor treated. Patients undergoing this process often migrate between physician providers in search of a definitive diagnosis of their troublesome symptoms. Interviewed physicians estimate that 30% of subclinical hyperthyroid patients have borderline-low TSH levels that present additional diagnostic challenges given the high degree of variation in existing definitions of normal, low, and borderline-low TSH levels (

 Evidence-based clinical guidelines for management of subclinical hyperthyroidism are lacking.

 Furthermore, modest excess of thyroid hormone can have detrimental effects on cardiac function, bone turnover, and cognitive function, and is associated with increased all-cause and cardiovascularrelated mortality,

 which may suggest that certain subclinical hyperthyroid patients may benefit from thyroid-regulating therapy.

Given the inherent complexities and diagnostic challenges outlined above, a patient may spend months being evaluated and treated by specialists other than endocrinologists, which may result in an increase in otherwise avoidable office visits, laboratory tests, and symptom management.

The most commonly used test for diagnosis of GD is the relatively expensive thyroid radioactive iodine uptake and scan (RIUS).

Thyroid RIUS tests are used to differentiate GD from other causes of hyperthyroidism and can be important for determining a course of action in patients who will undergo radioablation or surgical removal of the thyroid. Recently published guidelines have promoted the value of measuring anti-TSH receptor antibodies as an alternative way to diagnose GD when thyroid scintigraphy is not available or is contraindicated.

The US Food and Drug Administration recently cleared a commercially available reporter bioassay (Thyretain) that specifically detects TSI,

 which is the functional type of anti-TSH receptor that causes GD. Taken together, these reports suggest that TSI testing may provide a

more streamlined diagnostic algorithm that could reduce the dependence on thyroid RIUS, and substantially shorten the time to GD diagnosis.

The aim of the current study is to determine whether utilization of the TSI bioassay test will reduce overall costs of diagnosis to payers and shorten the time to diagnosis of GD. We use a budget impact model to assess the impacts of early TSI testing on costs and time to diagnosis.

The model was designed from a managed care payer perspective, specifically managed care organizations (MCOs) that have assumed financial risk for the care of their members. The target audiences are decision and policy makers of health plans, health systems, and other MCOs.

The model is used to estimate the impact of costs on the MCO’s annual budget. Analyses of cost per patient are estimated on a cost-per-diagnosis basis.

The model uses a theoretical population of 100,000 enrollees in an MCO. The percentage of patients in an MCO presenting for TSH testing for hyperthyroidism

is estimated at 14.6% per year and is based upon the percentage of the US population that is tested for TSH each year. This estimate was based on a published estimate that 45 million first-time TSH tests are performed in the United States each year per 300 million people

; 6.3% of patients were estimated to have low or borderline-low TSH levels based on primary research interviews. An occurrence rate of 0.644% for suspected GD was estimated based upon the average estimate that 70% of those with hyperthyroidism have GD (Table 1).

Laboratory and office visit costs associated with currently used and theoretical TSI test—based algorithms were calculated and compared (

) to determine direct cost savings. Misdiagnosis costs associated with current and TSI test-based algorithms were estimated and compared to determine additional cost savings (Table 2). Total costs to the payer were determined as the sum of direct and misdiagnosis costs and were compared between current and TSI algorithms to determine total cost savings to the payer (Table 2).

Cost savings discussed here are based on fee-for-service contracts only for Medicare, Medicaid, and commercial health plan enrollees, or covered lives. General findings also hold true for capitated contracts. Based upon known demographics of GD patients, a mix of plan types was assumed. The theoretical distribution of enrollees receiving TSH testing as percentage of total enrollment is: 20% Medicare, 18% Medicaid, and 62% commercial (Table 1). Reimbursement rates for Current Procedural Terminology (CPT) codes assigned to each laboratory test and office visit were obtained from the Centers for Medicare and Medicaid Services (CMS) 2010 fee schedules and the 2010B CMS physician fee schedule, which were used to estimate Medicare costs for each code.

For office visit—related costs, the average of non-facility and facility costs for each code was used. For RIUS test costs we used 75% of published hospital rate charges from a commercial payer to estimate the cost to Medicare.

 Based on industry averages, Medicaid reimbursement rates were estimated at 80% of Medicare reimbursement rates and commercial payer reimbursement rates were estimated at 120% of Medicare reimbursement rates.

The occurrence or prevalence rate for the percentage of patients that test low for TSH is based on the incidence rates for hyperthyroidism in the United States.

 The percentage of patients with subclinical versus clinical hyperthyroidism, percentage of patients with low to borderline-low TSH levels, and the misdiagnosis rates (Table 1) were established through in-depth, third-party interviews with practicing endocrinologists (detailed below). The percentage of hyperthyroidism patients with an underlying etiology of GD was estimated at 70% based on published estimates.

 De-identified records of thyroid function assays performed between January 1, 2007, and June 30, 2008, were analyzed by Management Science Associates, Pittsburgh, Pennsylvania, to identify

 International Classification of Diseases, 9th Revision, Clinical Modification 

) codes for conditions that are commonly misdiagnosed in patients with hyperthyroidism.

Five physicians were selected to provide expert feedback on the diagnostic process because they were familiar with the TSI test and were key opinion leaders in endocrinology. They were located in California, New York, New Jersey, and Florida. The interviews were conducted individually using a questionnaire to guide the discussions. Experts were asked to estimate the average time that it currently takes to arrive at a GD diagnosis; all 5 agreed, on average, that it was 2 to 3 months from the time the patient sees a primary care physician (PCP) to being diagnosed by an endocrinologist. They were asked to describe currently used diagnostic algorithms for hyperthyroidism and GD. The consensus was that the type of algorithm used can vary and depends upon whether patients have overt symptoms of GD. A variety of diagnostic algorithms were discussed and a representative range of these different algorithms was used in the model. When asked to estimate the percentage of patients that present with obvious symptoms of hyperthyroidism, the average response was 70% (range 65%-75%). We also asked physicians to estimate the percentage of patients that have GD and are misdiagnosed, what diseases GD is mistaken for, and how long it takes for a misdiagnosis to be identified. Three of the 5 physicians estimated the misdiagnosis rate at around 5%, while 2 estimated that it was close to 20% of patients. Because these estimates varied by more than 20%, we used the more conservative estimate of 5% in the model, which would lead to a more conservative estimate of cost savings. All physicians estimated that misdiagnosed patients spend 6 months undergoing treatments for the wrong disease before being correctly diagnosed with hyperthyroidism. Finally, we asked physicians how the TSI test could impact this rate of misdiagnosis, which was estimated at 85% based on the sensitivity and specificity of the TSI test. With the exception of the estimated rate of misdiagnosis, the answers varied less than 20%.

Clinical practice guidelines and in-depth primary research interviews with 5 key opinion leaders in endocrinology were used to identify and validate the mix of current GD diagnostic algorithms. Algorithms that capture current diagnostic practice were outlined for varying TSH results and clinical presentation (

). Hypothetical algorithms that include TSI testing were similarly defined in order to compare costs with current algorithms (

The model assumes that physicians presently utilize a set number of highly structured algorithms, and that they would utilize similarly structured algorithms as they adopt the TSI test. In clinical practice, however, physician behavior likely draws upon a wider variety of less formally structured algorithms; such behavior would likely continue even as they adopt the TSI test. The model uses non-weighted averages to determine costs for all current algorithms and TSI test-based algorithms for patients with clinical disease; in practice, some algorithms may be more heavily utilized than others. The model assumes that patients with clinically overt Graves’ disease do not have borderline-low TSH levels, always exhibit TSI, and that all patients with subclinical hyperthyroidism go through the diagnostic process. In the model, treatment costs for misdiagnosed patients are experienced at a constant rate; however, in front-line clinical practice, these costs will likely vary based on several factors including stage of disease progression, type and location of service provider, and terms of payer contracts.

Comparison of Direct Costs of GD Diagnosis: Current Versus TSI Algorithms

An Excel-based budget impact calculator model was designed to assess the cost of current GD diagnostic algorithms (Figure 1A) compared with the cost of theoretical algorithms utilizing TSI testing (Figure 1B).

The cost of each diagnostic algorithm shown in Figure 1A is driven by costs of the following: 1) laboratory tests (TSH, triiodothyronine resin uptake [T3RU]/thyroid hormone—binding ratio [THBR], FT4, FT3, thyroid receptor antibody, thyroid peroxidase antibody, and TSI), 2) RIUS test and interpretation, and 3) number of PCP and endocrinologist/specialist office visits (Figure 1, Table 1). For TSI diagnostic algorithms in subclinical hyperthyroid patients, a weighted average was used because endocrinologists predicted that 20% of subclinical GD patients would not exhibit detectable levels of TSI; for all other patient types, a flat average was taken across the diagnostic algorithms. A weighted average of the diagnosis cost for current and TSI algorithms was taken across patient types based on the percentages of clinical and subclinical hyperthyroid patients with low or borderline-low TSH (Table 1). The estimated GD diagnosis cost associated with current algorithms was compared with the estimated GD diagnosis cost associated with TSI algorithms to establish overall cost savings.

Comparison of GD Misdiagnosis Costs: Current Versus TSI Algorithms

De-identified claims data were used to quantify the most common 

codes and related misdiagnoses in GD patients (Table 1). The model assumes that misdiagnosed patients are treated for 6 months before receiving an accurate diagnosis. Costs associated with treatment of each of these conditions per year were calculated using published literature.

 The number of patients impacted by misdiagnosis was based on the expert interviews and indicated misdiagnosis rates of 5% and 0.75% for currently used algorithms and TSI algorithms, respectively. Costs of misdiagnosis were calculated per patient and multiplied by the number of patients with low to borderline-low TSH per year in a population of 100,000 enrollees to determine annual total costs associated with misdiagnosis (Table 1).

Total costs to the payer were determined by adding the total direct costs and the costs of misdiagnosis. The impact on a budget of 100,000 enrollees was determined by using the occurrence rate to determine the number of patients undergoing testing for hyperthyroidism.

Time to Diagnosis: Current Versus TSI Algorithms

Articles

WEB EXCLUSIVE

An evidence-based managed care budget impact model shows that incorporating TSI testing into existing Graves%u2019 disease hyperthyroidism diagnostic algorithms reduces costs and shortens time to diagnosis.

TSI Assay Utilization: Impact on Costs of Graves' Hyperthyroidism Diagnosis

 To evaluate the impact of paperless provider credentialing in a multi-state managed care organization.

 Implementation took place from June to October of 2009. A total of 6220 providers were credentialed during this study period and selected for analysis. We used an interrupted time-series design, centered the data on implementation, and then compared efficiency rates for the 14 weeks before implementation of paperless credentialing to efficiency rates for the 14 weeks after implementation.

 The absence of a control group and a relatively short observation period are potential threats to validity.

 The main unit of analysis was the provider credentialing file. We compared quality review pass rates, processing time (in days), and processing cost per file before and after paperless credentialing implementation.

 The percentage of files passing quality review increased from 83% to 92%. The turnaround time for the credentialing process dropped from 53 calendar days to 36 calendar days.

 Implementation of electronic credentialing appears to significantly improve processing efficiency. Indeed, credentialing seems to be a particularly promising area for adoption of electronic records in an administrative setting.

	Healthcare provider credentialing is required to limit risk and ensure quality of provider networks. Providers cannot be contracted until credentialing is completed. The bigger the network, the more important it becomes to use an efficient and effective credentialing process. In this case study, implementation of a paperless credentialing process resulted in the following outcomes:

		Improved quality scores: 83% before, 92% after.

		Reduction in file processing times: 53 days before, 36 days after.

	Compared with other sectors in the US economy, the healthcare industry has been slow to adopt information technologies.

 Recent estimates suggest that less than a quarter of US hospitals have even a moderate level of organizational paperless record utilization.

Yet electronic health records have been shown to improve communication, collaboration, and efficiency.

 Transitioning to paperless records has also been shown to reduce costs

 and to improve document fidelity within health organizations.

 The existing literature on electronic medical records is largely related to clinical use and clinical outcomes, and thus is largely unrelated to fully administrative processes such as the one presented in this case study, which is not clinical in nature.

	In this case study, we present a productivity and efficiency analysis of the implementation of a paperless healthcare provider credentialing system within a multi-state managed care organization. All primary care providers, most physician/osteopathic medicine specialists, and a wide array of ancillary providers (eg, nurse practitioners, physical therapists, podiatrists, or social workers) require credentialing, as do several types of health facilities (eg, hospitals and skilled nursing facilities). Every healthcare organization (eg, insurance agency, hospital, or licensing board) must verify provider credentials before contracting with, employing, or directing organization members to a provider. Consequently, there is a very large need for the credentialing process to be as expedient as possible, in order to ensure that members and patients of each respective organization have appropriate access to healthcare providers.

	Credentialing typically involves various provider record checks (ie, state licensure, board certification, education, etc) and often requires extensive documentation and frequent communication with providers to clarify or verify these details. This process must be completed every 2 to 3 years while the provider remains contracted with an organization. Credentialing is entirely document- and process-driven, and the industry generally relies on paper records rather than electronic records. In addition, credentialing regulations vary by state and frequently change. Consequently, credentialing is often a time-consuming and expensive process. Each day that a credentialing file remains in process is another day of incurred cost, so any reduction to file processing time that results from the implementation of an electronic records process leads to a reduction in the cost of credentialing for the organization.

	Molina Healthcare is a managed care organization that serves more than 1.5 million beneficiaries in multiple states. The Spokane, Washington, office houses the corporate credentialing office, and processes credentials for approximately 300 to 500 providers per week. From June to October 2009, the office implemented a paperless credentialing process, with implementation staggered over 8 different geographic business regions. This was necessary due to the large scope of the implementation and the impracticality of slowing or interrupting business for the entire department and risking a critical work flow failure. While regionalization led to variable volumes for each implementation segment, the regionalization of implementation would not have had a significant impact on outcome variables because the core credentialing criteria and implementation plan utilized was identical from region to region. To prevent an immediate production failure, credentialing employees were trained to use the new paperless healthcare provider credentialing system (Laserfiche software) prior to first production use.

	Various data elements were tracked for each file processed by each employee in Molina Healthcare’s credentialing database (CACTUS software), and these data were pulled ad hoc from the database for analysis in this study. We limited our analysis to file processing time and file quality for files completed during the 14 weeks prior to implementation and 14 weeks post-implementation, with data for each file centered around the implementation date respective to that file (due to the staggered implementation schedule). There were a total of 8068 files that fell into the study period before selection criteria were applied. For the data categorized into the 14 weeks post-implementation, only files that were started and completed in a paperless format were included; any paper-based files that were completed during the 14 weeks post-implementation were removed from analysis. Data for the week of implementation were removed due to the lack of inherent score validity during that week. Records processed by new employees, departing employees, and employees on leave for 1 or more weeks during the study period were omitted from the analysis to control for any existing learning curve and any vacationrelated productivity backlog. Records that were only partially completed would not have been submitted for quality review and were not included in the analysis. A total of 6220 provider files were selected for final analysis after these criteria filters were applied; 6080 of these records were unique provider files, while 140 provider records were included twice in the analysis due to necessary early recredentialing.

	The following 2 measures were used to track credentialing productivity:

 Average percentage of files meeting quality review standards. Each file was coded as a pass or fail by a quality lead upon review. For analysis, the number of files completed that passed quality review was divided by the number of files completed overall. This was analyzed on a weekly pre-/post-implementation\ basis to study chronological trends, and on an overall pre-/post-implementation basis to compare pre/post means.

 Average number of days needed to complete the credentialing process. This variable was measured in calendar days from the day an application was received in the corporate credentialing department to the time a file was deemed as passing the quality review process. For analysis, file processing times were averaged for all files that were deemed complete in the quality review process. This was analyzed on a weekly pre-/post-implementation basis to study chronological trends, and on an overall pre-/post-implementation basis to compare pre/post means.

 The proportion of files passing quality review rose from 83% (of 3149 files reviewed) during the 14 weeks pre-implementation to 92% (of 3071 files reviewed) during the 14 weeks post-implementation. During weekly pre-implementation analysis, the lowest score was 74% (of 205 files reviewed) and the highest score was 89% (of 250 files reviewed). During post-implementation analysis, the low score was 86% (of 288 files reviewed) and the high score was 94% (of 284 files reviewed). 

 Average file-processing time dropped from an average of 53 calendar days (for 3149 files reviewed) during the 14 weeks pre-implementation to an average of 36 calendar days (for 3071 files reviewed) during the 14 weeks post-implementation. During pre-implementation as viewed on a weekly basis, the highest score was 62 days (for 205 files reviewed) and the lowest score was 49 days (for 228 files reviewed). During post-implementation as viewed on a weekly basis, the highest score was 43 days (for 256 files reviewed) and the lowest score was 29 days (for 223 files reviewed). 

	This study offers provocative, though not definitive, evidence that paperless credentialing improves efficiency. The outcome of greatest interest to business is reduced file-processing time because lags in credentialing can delay provider enrollment, which, in turn, impacts financial outcomes. Provider payment rates are based on contracted and credentialed provider status, and each day a file is in the credentialing process results in additionally incurred man-hour expense. Molina’s credentialing unit is charged with building and maintaining a provider network that is of sufficient quality and that meets geographical access needs. It is also responsible for ensuring that the cost, efficiency, and expediency of the credentialing process meet business performance requirements. We observed a substantial drop and then a slow increase in file-processing time immediately after paperless implementation, stabilizing several weeks later at a level that was significantly lower (in an operational sense) than the pre-implementation file-processing time. A similar trend was observed in the quality measure, which also showed less week-to-week variability post-implementation due to the increased standardization of documentation that came with the paperless transition. These findings should help allay operational concerns about use of electronic records for credentialing.

	The internal validity of additional studies in this area would be improved by a more lengthy observation period and a comparison group. While the quality measure does give a level of confidence in performance improvement—which leads to reduced double work and lowered departmental costs&mdash;the use of a pass/fail quality process could be revised to result in a “score per file,” which would allow for analysis of the impact of paperless implementation on quality outcomes in a more complex way. Another supporting variable that would support study outcomes in future studies would be a cost-per-file variable. It was not possible to construct a cost analysis that aligned with the study methodology due to the monthly structure of financial budgets (which did not draw out the distinction between paper-based labor and paperless labor). It would be beneficial in future studies to include a full-cost study based on implementation structure and study methodology. It is worth noting that a chronological file-processing cost analysis was completed for the implementation presented in this study, which was based on monthly financial budget data. This cost assessment removed all costs not directly related to file processing and removed costs for all employees whose production was not included in the study. This assessment showed that the cost per credentialing file for the whole department before any paperless implementation was $180, versus $108 after all paperless implementation.

	For any organization considering such a transition, it is important to note a few of the operational concerns that may arise during paperless implementation. Any organization considering a full paperless implementation should be vigilant of the impact of introducing fully electronic data. The use of electronic data presents several opportunities for potential process automation; however, it also increases the reliance on computer systems and there may be new interoperability or contingency planning issues to consider. Molina obtains approximately 65% of its data from standardized and centralized data sources (ie, the Council for Affordable Quality Healthcare). The opportunity to retrieve information in an automated and batch fashion from these sources became a realistic possibility upon implementation of the paperless software, which greatly reduced the hours of manual labor needed to initiate the credentialing process on the front end. This type of work flow reorganization should be considered where data exchange between different sources exists (ie, software programs, data vendors, etc). This is perhaps one of the greatest opportunities in the paperless environment for administrative simplification.

	Study was completed on corresponding author’s time as part of a master’s program at Washington State University. Molina Healthcare was the source of data and location of implementation, and had final approval of publication of manuscript.

 From Health Policy and Administration, Washington State University (JK, JSC, GJS, RJB), Spokane, WA; Molina Healthcare (RJB), Spokane WA.

 Mr Boe reports employment with Molina Healthcare. The other authors (JK, JSC, GJS) 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, RJB); acquisition of data (RJB); analysis and interpretation of data (JK, JSC, GJS, RJB); drafting of the manuscript (JK, JSC, GJS, RJB); critical revision of the manuscript for important intellectual content (JSC, GJS, RJB); statistical analysis (JK, RJB); administrative, technical, or logistic support (GJS, RJB); and supervision (JK, GJS, RJB).

 Ryan J. Boe, MHPA, 620 W. Montgomery Ave, Spokane, WA 99205. E-mail: ryanjboe@gmail.com.

	1. Mills TR, Vavroch J, Bahensky JA, Ward MM. Electronic medical record systems in critical access hospitals: leadership perspectives on anticipated and realized benefits. 

	2. Morris S. The path to a paperless practice. 

Institution of paperless credentialing is analyzed on a pre-/post-implementation basis to understand the impact on business and productivity.

Implementation of Paperless Credentialing in a Multi-State Managed Care Organization

 Many patients with depression do not respond to first-line antidepressant therapy and may require augmentation with another concurrent treatment such as a second antidepressant, a stimulant, a mood stabilizer, or a second-generation antipsychotic (SGA). The objective of this study was to examine the relationship between patient cost-sharing and the use of augmentation among a sample of commercially insured patients.

Retrospective observational study of adult patients diagnosed with depression and receiving antidepressant therapy (n = 48,807).

 Logistic regression models estimated the likelihood of augmentation as a function of patient cost-sharing amounts. An alternative-specific conditional logit model of the likelihood of each augmentation class, varying the cost-sharing prices faced for each class, was also estimated. All models controlled for sociodemographic characteristics, physical and mental comorbidities, health plan type, and year of index antidepressant therapy initiation.

 The range of mean copayments paid by patients for augmentation therapy was from $27.05 (antidepressant) to $38.81 (SGA). A $10- higher cost-sharing index for all augmentation classes was associated with lower odds of augmentation (adjusted odds ratio = 0.85; 95% confidence interval 0.79-0.91). Doubling the costsharing amount for each augmentation class was associated with a smaller percentage of patients utilizing each class of augmentation therapy.

Employers and payers should consider the relationship between cost-sharing and medication utilization patterns of patients with depression.

	In a sample of adult patients with depression who were undergoing antidepressant therapy, higher patient cost-sharing is associated with a lower likelihood of antidepressant treatment augmentation. Higher cost-sharing amounts for each augmentation class were associated with a smaller percentage of patients utilizing each class of augmentation therapy.

		Our findings add to the evidence that cost-sharing not only impacts treatment decisions for depression from the time of initial therapy but also follow-on treatment  pathways for those not achieving remission.

		 Employers and payers should consider the implications of higher cost sharing on the utilization patterns of patients with depression.

Depression is among the most common psychiatric disorders in the United States. Using data from the US National Comorbidity Survey Replication conducted from February 2001 through April 2002, Kessler et al (2005) found the 12-month prevalence of major depressive disorder to be 6.7 percent.

 Consistent with the high rate of depression in the population, antidepressants are among the most frequently prescribed medication classes in the United States.

Patients with depression may not respond to first-line antidepressant therapy; even adequate trials of antidepressants often fail to achieve remission of symptoms. The Sequenced Treatment Alternatives to Relieve Depression (STAR*D) study reported a remission rate for first-line therapy (citalopram) of up to 32.9%, depending on the definition of remission.

 For patients who do not respond to first-line antidepressant therapy, treatment options include remaining on a single antidepressant (monotherapy) or augmenting antidepressant treatment with another concurrent treatment such as a second antidepressant, a stimulant, a mood stabilizer, or a second-generation antipsychotic (SGA).

While treatment decisions are primarily based on clinical considerations, patient cost-sharing (ie, the price to the patient) may also be a consideration when deciding which treatment course to pursue. Previous research has demonstrated that higher patient prescription drug costsharing amounts are associated with lower levels of prescription drug utilization and expenditures.

While one rationale behind imposing higher prescription drug cost-sharing is to steer patients away from low-value drugs and preventing medication overuse, previous studies have shown that higher cost-sharing amounts are also associated with lower initiation rates of high-value medication therapies for chronic illnesses.

The largest financial impact on the patient in the decision to remain on a single therapy, or to attempt augmentation, is the choice of continuing to pay 1 cost-sharing amount or to bear an additional cost-sharing amount when adding a treatment. To our knowledge, no studies have addressed the effects of cost-sharing on treatment augmentation patterns in patients on antidepressant therapy.

The objective of this study was to examine the relationship between patient cost-sharing and the use of augmentation strategies among a large sample of commercially insured patients with depression on antidepressant therapy. We examined the effects of cost-sharing on any augmentation with a mood stabilizer, SGA, stimulant, or antidepressant. In addition, we estimated the relationship between changes in cost-sharing within each class on the share, or percentage, of patients utilizing each augmentation class.

This retrospective analysis used data from the Thomson Reuters MarketScan Commercial Claims and Encounters (CCAE) Database, which contains the healthcare experience of tens of millions of individuals annually who have commercial health insurance provided primarily by large self-insured employers. The CCAE Database includes detailed spending and utilization data for healthcare services performed in both inpatient and outpatient settings, covered by a variety of plan designs, including preferred provider organizations (PPOs), point of service plans, indemnity plans, and health maintenance organizations (HMOs). Medical claims are linked to outpatient pharmacy claims and enrollment data using unique enrollee identifiers. No institutional review board approval was required, because the database meets criteria for a limiteduse data set in compliance with the Health Insurance Portability and Accountability Act of 1996 (HIPAA).

Patients were selected from the database by finding the first antidepressant claim for each patient during the study time frame (January 1, 2005, through December 31, 2008) with a 1-year period of continuous enrollment prior to the antidepressant fill (

). Augmentation of antidepressant treatment is typically not recommended for patients for at least 8 weeks after initiation of antidepressant therapy.

Conservatively, to allow for partial response, augmentation was required to be at least 4 months (16 weeks) after the initial antidepressant, allowing two 8-week antidepressant trials.7 The index date was established as the date 4 months after the initial antidepressant, and each patient had a minimum follow-up period of 1 year (52 weeks) following the index date.

We selected patients who were 18 to 64 years of age at the time of the first antidepressant prescription claim and had at least 2 medical claims (outpatient or inpatient services) with a depression diagnosis (

International Classification of Diseases, Ninth Revision, Clinical Modification

]: 296.2x, 296.3x, 300.4, 309.0, 311) during the study period, which consisted of the year preceding the initial antidepressant claim and the 16 months following. Patients were excluded from the study if a prescription for one of the augmentation classes appeared prior to the index date (n = 5672) or if they had evidence of electroconvulsive therapy (n = 5) or a prescription claim for clozapine or a fixed-dose combination of antidepressants and SGA (n = 135). After these exclusion criteria were applied, patients were also excluded if they had a medical claim with another diagnosis wherein one of the augmentation strategies might be indicated. These included dementia, schizophrenia, delusional disorder, psychoses, pervasive development disorder, mental retardation, cerebral degenerations, Parkinson’s disease, senility, manic depression, bipolar disorder, or major depressive disorder with psychotic features (n = 688). Patients were also excluded if the first augmentation prescription did not overlap with an antidepressant prescription for at least 30 days (allowing 6 gap days) or if prescriptions for medications within 2 different augmentation classes appeared on the date of the first augmentation (n = 4823). We found 48,807 patients meeting the study criteria of antidepressant medication treatment and fulfilling the diagnostic and other criteria.

The first outcome measure was the use of any augmentation therapy in the post-index time frame, and any augmentation was recorded as a yes/no indicator variable in the year following the index date. Augmentation of the initial antidepressant therapy was defined as at least 30 days overlap (with a 6-day gap) post-index with any 1 of the following classes: SGA, mood stabilizer, stimulant, or a second antidepressant.

The first class of augmentation (SGA, mood stabilizer, stimulant, concomitant antidepressant, or antidepressant monotherapy [no augmentation]) that the patient filled in the year following the index date was recorded.

The key explanatory variable was the patient prescription drug cost-sharing index

representing the prices faced by each patient within each employer/health plan combination for a fill within each augmentation class, regardless of whether they augmented or not. Cost-sharing is represented by a price index calculated as a weighted average of the prices of each medication, as patients are likely to respond to the prices of each medication alternative they face, even if they do not fill a prescription within that class. Class-specific (eg, mood stabilizer or SGA medication), plan-level cost-sharing amounts were calculated as a weighted average of brand-name and generic copayments (

) with weights based on overall utilization (a market-basket approach), 

 = 1. In the case of SGA medications, no generic medications existed at the time, so the class-specific index was equal to the brand-name cost-sharing amount. In addition to the class-specific measure, a plan-level index combined a weighted average of the brand and generic cost-sharing amounts for all classes (with the utilization weights summing to 1).

Other explanatory variables included sociodemographic characteristics consisting of age, gender, employee status (vs spouse or dependent), rural (vs urban) residence, United States Census region, and median household income by zip code of residence (from the US Census). Type of health plan (eg, HMO, PPO) was also included. Health status variables were measured during the year prior to the index date. The Charlson Comorbidity Index (CCI) is an aggregate measure based on diagnoses associated with 19 conditions.

 Because the CCI does not capture mental health conditions, the number of Psychiatric Diagnostic Groupings (PDGs)

 was also included. The PDGs are identified by 

 diagnosis codes and include comorbid mental health conditions such as alcohol use disorders and other substance use disorders. A dichotomous flag indicating medication switching when at least 2 different antidepressant monotherapies were filled during the 4-month period following the initial antidepressant prescription was included, which may signal treatment resistance to the initial antidepressants. Also included were indicator variables denoting the year the first observed antidepressant was filled.

Two sets of models were estimated for the outcome of any augmentation. First, a logistic regression was used to model the odds of any augmentation over a 1-year period after the index date. Based on the model results, we calculated the effects of higher cost-sharing on the likelihood of any augmentation and the price elasticity of the probability of augmentation. Models included cost-sharing variables, sociodemographic characteristics, health plan type, health status variables, and time. We also included a fixed effect for each employer (firm) to account for all time-invariant employer-specific confounders in the analysis.

For the analysis of the outcome on augmentation class, we estimated an alternative-specific conditional logit model (McFadden’s choice logit) of the type of augmentation (augmentation class) utilized within 12 months of the index date.

The alternatives in this model were the 5 augmentation strategies (none, SGA, mood stabilizer, stimulant, antidepressant). This model is similar to the multinomial logit; however, in the alternative-specific conditional logit model, the cost-sharing  amounts were varied with each augmentation class (alternative-specific characteristics). Other covariates remained the same with each augmentation class (ie, sociodemographic, health plan, health status, time, and firm fixed effect). Based on the model results, we calculated the effects of higher costsharing for each augmentation alternative on the probability that each augmentation alternative is selected.

 Using results from the alternative-specific conditional logit model, we estimated the effects of a change in the cost-sharing index for each augmentation class, varying the prices 1 class at a time, on the share of patients using the class. Using this information, we calculated the own price elasticity of share (ie, the percent change in the share of patients using the class/percent change in cost-sharing for the class).

Of the 48,807 patients meeting the study criteria, 10.2% (4984) augmented therapy within 1 year of the index date (ie, 4 months after the first antidepressant fill). Most augmented with an antidepressant (7.8%), followed by a mood stabilizer (1.3%), and then SGAs and stimulants (both 0.6%).

Patient characteristics by augmentation class are displayed in 

.The average age in each cohort was about 40 years, although patients augmenting with SGAs were the youngest on average (38.4 years), and patients augmenting with a mood stabilizer were oldest (42.3 years). Patients augmenting with a mood stabilizer were also most likely to be female (67.1%) and patients augmenting with an SGA were most likely to be male. About 40% of patients had an initial antidepressant fill in 2005, another 40% had an initial fill in 2006, and the remainder had an initial fill in 2007.

Health status varied across the cohorts, with patients augmenting with mood stabilizers having the highest CCI score (0.37) and the lowest number of PDGs (0.74). Conversely, patients on antidepressant monotherapy had the lowest CCI (0.20) and the highest number of PDGs (0.80). Almost one-fifth of patients augmenting with SGAs had 2 or more antidepressant agents in the 4 months after the initial antidepressant fill, while half that amount (10.1%) of patients on antidepressant monotherapy filled a prescription for 2 or more antidepressant agents during the same time.

The mean cost-sharing index varied with the medication class (Table 1) and the smallest amount was for antidepressants, at $27.05 per fill; SGAs were the highest, at $38.81 perfill. The overall cost-sharing index, combining all classes as a weighted average, was $28.67 per fill.

Results from the logit model of any augmentation showed that as the cost-sharing index rose, the likelihood of any augmentation decreased (

). A $10-higher cost-sharing index for all augmentation classes was associated with 15% lower odds of augmentation (adjusted odds ratio [OR] = 0.85, 95% confidence interval [CI] 0.79-0.91) (Table 2). When converted into the effects on the probability of any augmentation, a $10-higher cost-sharing index was also associated with 1.5-percentage- point lower (

 <.01) probability of any augmentation. This translates to a price elasticity of the probability of any augmentation of —0.44. Figure 2 displays the predicted probability of any augmentation at various cost-sharing levels.

Results from the alternative specific conditional logit model showed that higher cost-sharing was associated with lower odds of augmentation (

). While prices for each augmentation class were entered separately into the model, a single marginal effect of cost-sharing was obtained, representing the association between cost-sharing and augmentation. We found that higher cost-sharing was associated with lower odds of augmentation (adjusted OR = 0.85, 95% CI 0.75-0.91).

We found that a doubling in the price, or 100% higher price, of antidepressants was associated with a 41% smaller share of patients utilizing antidepressants as an augmentation class (elasticity estimate of —0.41). Also, a doubling in the price of SGAs led to a 63% smaller share of patients utilizing SGAs (elasticity estimate –0.63);  a doubling in the price of mood stabilizers or stimulants was associated with a slightly smaller share of patients utilizing mood stabilizers (49% smaller, elasticity –0.49) or stimulants (55% smaller, elasticity –0.55), respectively.

This study contributes new evidence to the literature regarding the impact of patient cost-sharing on utilization of medications to augment antidepressant therapy among patients with depression. Among a population of adults with depression who have employer-sponsored health insurance, higher cost-sharing was linked to lower rates of utilization of classes of medication used to augment antidepressant therapy. Specifically, higher prescription drug cost-sharing was associated with lower rates of any augmentation. Importantly, higher cost-sharing amounts for each augmentation class were associated with a smaller percentage of patients utilizing each class of augmentation therapy.

Higher patient cost-sharing is associated with a lower likelihood of treatment augmentation in patients with depression who are treated with antidepressants.

Impact of Cost-Sharing on Treatment Augmentation in Patients With Depression

 The Cesarean section (CS) rate in Taiwan has exceeded 30% since 2000. To lower the CS rate, the Bureau of National Health Insurance in Taiwan raised the payment for vaginal delivery (VD) in May 2005, and also increased the insured’s copayment for elective CS in May 2006. This study clarifies the influences of these financial incentives, and explores whether the 2 policies lowered the CS rate.

 The materials used are birth cases obtained from a systematic sampling of the original inpatient claim data in the National Health Insurance research database between 2003 and 2007. The empirical analysis of this study groups the birth data into 4 types: VD, unplanned CS, planned CS, and elective CS.

The 4 delivery types represent the dependent variable. A multinomial logistic regression model was adopted as the empirical method. The policy changes, hospital attributes, and insured status were considered independent variables.

 Results indicate that the supply-side policy change in May 2005 reduced the number of CS cases. However, the policy effect was not very significant, and the CS rate decreased mainly because of planned CS cases. The demand-side policy change in May 2006 did not lower the rate of elective CS.

The results imply that the financial incentives were not the main consideration for both the demand and supply sides. To encourage more VDs and lower the CS rate, the authorities could consider mechanisms other than adjusting the payment or changing the copayment.

	The objective of policy changes in May 2005 and May 2006 was to lower the Cesarean section (CS) rate in Taiwan by providing financial incentives for supply and demand sides.

		This study clarifies the influences of financial incentives for both the demand and supply sides, and explores whether the 2 policy changes lowered the CS rate.

		The results indicate that the financial incentives were not the main determinant in choosing the delivery mode.

		To encourage more vaginal deliveries and lower the CS rate, the authorities could consider mechanisms other than adjusting the payment or changing the copayment.

	Since 2000, the annual Cesarean section (CS) rates in Taiwan have exceeded 30%.

 By comparison, in the 2000s, the CS rate in the US was 25% to 30%, but in some European countries, such as England, Germany, and Denmark, it was lower. Obviously, the CS rate in Taiwan has been higher than in many other countries for decades.

 To lower the CS rate, the Bureau of National Health Insurance (BNHI) in Taiwan adopted 2 important policies to encourage more vaginal deliveries (VDs). First, they raised the payment to healthcare providers for VD in May 2005. Second, they increased the insured’s copayment for elective CS in May 2006.

	Taiwan’s National Health Insurance (NHI) system adopted a case payment mechanism, the diagnosis-related groups system, to reimburse hospitals for deliveries (including both VD and CS). In addition, the BNHI regulates the medical indications of CS. The insured’s copayment for childbirth is zero if the delivery mode is VD or CS-recommended because of medical indications. However, a number of pregnant women ask for CS, even though their clinical conditions do not coincide with the medical indications of CS.

 These cases are categorized as “elective CS.” The CS cases recommended because of medical indications are categorized as “indicated CS.” The payment for indicated CS cannot be applied directly to elective CS.

	Prior to May 1, 2005, the provider payment for “indicated CS” was approximately US $911 to $1132 but the payment for VD was only US $506 to $609. The payment rate for “indicated CS” was nearly 80% greater than that for VD. On May 1, 2005, the BNHI raised the payment rate for VD by approximately 80%, but sustained the rate for CS. This payment adjustment raised a typical obstetrician’s income by approximately 17%. The payment of VD and indicated CS for healthcare providers has been equal since that time. The cost of performing a CS is often higher than that of performing a VD. The BNHI hoped that the May 2005 policy change would provide the supply side with the financial incentive to recommend VD.

	The BNHI also hoped to provide the demand side with a financial incentive to accept VD through a policy change in May 2006. Before May 2006, the provider payment for elective CS was equal to that for VD. If laboring women asked for elective CS, they had to pay the hospital the difference between the rates for indicated CS and VD, and this amount was equivalent to the insured’s copayment for elective CS. Before May 2005, the copayment for elective CS was US $404 to $523, and that was roughly 1.4% to 1.8% of the average annual income per household. However, because of the payment adjustment in May 2005, the rate difference between indicated CS and VD, which was the copayment for elective CS, became zero between May 2005 and May 2006. To expand the insured’s copayment for elective CS, the BNHI set a lower payment rate for hospitals in May 2006. This policy change reduced the payment rate for elective CS to US $506 and required laboring women to pay the rate difference between indicated CS and elective CS. This means that the insured’s copayment for elective CS had been increased to US $475 to $697, which was approximately 1.5% to 2.2% of the average annual income per household. 

 lists the details of payment and copayment for different delivery modes before and after these 2 policy changes.

	Reviewing the literature on the high CS rate in Taiwan revealed several causes for elective CS: preferring an auspicious delivery time, arranging delivery time easily, avoiding delivery pain, worrying about the influence of VD on sex life, suggestions from relatives, and reimbursement of supplemental insurance. Among these causes, the financial burden or copayment of elective CS did not seem to have significant influence on the decision of the demand side.

The evidence from Australia and Norway also showed that maternal requests, delivery experience, and fear of a trying labor might make an important contribution to elective CS.

 However, decisions on delivery mode are also correlated with recommendations from physicians. Furthermore, policy changes, such as the payment for childbirth, might have affected the physicians’ financial incentives and their recommendations.

	Other researchers have also verified the effects of financial incentives and personal motives by physicians practicing CS. Gruber and Owings explored the birth data of different states in the United States from 1970 to 1982. During this period, the birth number fell by 13.5%, and the CS rate increased significantly. They indicated that obstetricians might have substituted CSs for some VDs to balance off their income loss because payment for CS was higher than that for VD.

 Moreover, Gruber et al and Grant confirmed that a higher payment for CS would induce a greater increase in the CS rates.

 Spetz et al examined the likelihood of CSs being performed at particular times of the day. They found that physicians’ time preference and financial incentives affectedtheir decision to perform a CS.

 displays the underlying motives for the supply and demand sides before and after the 2 policy changes in May 2005 and May 2006. Compared with the period before May 2005, hospitals or physicians now have the financial incentive to recommend VD. However, the financial incentives for the demand side are slightly complex. Between May 2005 and May 2006, the insured’s copayment for elective CS was zero. In this period, the demand side had the financial incentive to request elective CS. Before May 2005 and after May 2006, laboring women had to pay the copayment for elective CS, and the demand side may have been less likely to ask for elective CS. Furthermore, both the supply and demand sides had the convenience incentives to perform CS or ask for elective CS because the timing of CS could be arranged easily.

	Before and after the 2 policy changes in May 2005 and May 2006, the convenience incentives for the supply and demand sides remained unchanged. However, some changes occurred in the financial incentives for both sides. If the financial incentives were the important determinants for the decisions of physicians and women in labor, this study proposes the following 2 hypotheses: First, compared with the period before May 2005, indicated CS would decrease, but elective CS would increase between May 2005 and May 2006. The overall CS rate might not necessarily be reduced. Second, relative to the period before May 2005, indicated CS and elective CS would decrease after May 2006.

	The objective of both policy changes in May 2005 and May 2006 was to lower the CS rate in Taiwan by providing financial incentives for supply and demand sides. This study clarifies the influences of financial incentives on determining the delivery mode, and explores whether the 2 policies lowered the CS rate.

	The empirical materials are 51,085 birth cases obtained from a systematic sampling of the original inpatient claims data in the Taiwan NHI research database between 2003 and 2007. The data set contains 5% of inpatient admission records, extracted by systematic sampling method on a monthly basis, together with the related details of inpatient orders. The registry data of the insured and contracted hospitals have been adopted together.

	The birth sampling data from 2003 to 2007 are grouped into 4 types, which are VD, unplanned CS, planned CS, and elective CS. Both types of unplanned CS and planned CS are considered indicated CS, which was recommended by obstetricians according to the medical indications of CS. Following the study by Henry et al, the cases of indicated CS cord prolapse, fetal distress, long labor, disproportion, obstructed labor, and failed induction of labor are categorized as unplanned CS.

International Classification of Diseases, Ninth Edition, Clinical Modification 

) codes of unplanned CS included 653, 652 (except 652.1, 652.2, 652.5), 656.3, 659.0, 659.1, 660, 661 (except 661.3), 662, and 663.0. Other indicated CS cases are categorized as planned CS. The 4 delivery types are defined as the dependent variable. The multinomial logistic regression model was adopted as the empirical method and the standard errors were adjusted at the hospital level to account for the correlation of the data.

	Because the BNHI adopted 2 policies in May 2005 and May 2006 to lower the CS rate, the period from 2003 to 2007 could be divided into 3 segments by the 2 time points of policies. Period I represents the segment of 28 months before May 2005. Period II represents the segment of 12 months between May 2005 and May 2006. Period III represents the segment of 20 months following May 2006. Period I served as the benchmark, and 2 dummy variables (also known as indicator variables), Period II and Period III, were adopted to show the difference between those periods and Period I.

	This study also controls the influences of hospital attributes, the clinical conditions, and the insured status. Because all medical institutes can be classified according to the scale, there are 4 types of institutes: medical center, regional hospital, local hospital, and clinic. Three dummy variables, Medical Center, Regional Hospital, and Local Hospital, were adopted to show how they differed from the clinics in the CS rates. In Taiwan, the ownership of a medical institute is public, private, or juristic person. Public institutes are operated by the public sector, whereas private and juristic-person institutes are not. The owner of a private institute is a natural person, and the owner of a juristic-person institute is a juristic person (legal person). Two dummy variables, Public and Juristic-Person, were adopted to test how the public or juristic- person institutes differ from private institutes. The location of the medical institutes was also considered. According to the urbanization level, the locations of all institutes are categorized into 3 types: 1) Metropolitan Area, 2) Second Metropolitan Area, and 3) Rural Area. Two dummy variables of Metropolitan Area and Second Metropolitan Area were adopted to examine whether CS rates in metropolitan areas were higher than those in rural areas.

	Furthermore, the pregnant women might be registered in the NHI through their employers (or their occupational unions) or be enrolled as dependents of their employed spouses (or direct blood relatives). Therefore, 1 dummy variable, Dependent, was adopted to show the effects of the insured’s status. According to the records of inpatient claims and the insured, the age of each laboring woman was calculated and considered as being an independent variable in the regression analysis. Because obstetricians usually recommended the delivery method based on clinical conditions, a dummy variable of High Risk was considered to identify whether the clinical conditions of a laboring woman satisfied the characteristics of high-risk pregnancy. Following the studies by Lin et al and Han and Lien,

 this analysis also includes 2 dummy variables: 1) Weekend or Holiday and 2) Ghost Month. The former identified whether the birth date was a weekend or a holiday; the latter identified whether the birth date belonged to the ghost month. Ghost month is the Chinese lunar month of July. Many Taiwanese believe that the door of hell opens and wandering ghosts visit the human world during this month. Thus, ghost month is discouraging for hospitalization and elective surgery.

	The numbers of the sampling birth cases in periods I, II, and III are 24,624, 9958, and 16,503, respectively. In Period I, the case numbers of VD, unplanned CS, planned CS, and elective CS were 16,524, 2930, 4716, and 454, respectively. In Period II, the case numbers of VD, unplanned CS, planned CS, and elective CS are 6610, 1186, 1997, and 165, respectively. In Period III, the case numbers of VD, unplanned CS, planned CS, and elective CS are 10,939, 1978, 3229, and 357, respectively.

 shows the results of the multinomial logistic regression for 4 delivery modes, with VD the base group. The odds ratios of Medical Centers, Regional Hospitals, and Local Hospitals are less than 1 for the 3 CS types. The probability ratio of unplanned CS versus VD in medical centers is approximately 37% smaller than the ratio in clinics. The probability ratio of planned CS versus VD in regional hospitals is roughly 17% smaller than that in clinics. Regarding hospital ownership, the coefficients of Public and Juristic-Person for unplanned CS are positive, and those for planned CS are negative. The probability ratios of unplanned CS versus VD in public and juristic-person institutes are 21% to 24% greater than the ratio in private institutes. The probability ratios of planned CS versus VD in public and juristic-person institutes are 11% to 15% smaller than the ratio in private institutes. For elective CS, the odds ratios of Metropolitan Area and Second Metropolitan Area are 2.754 and 3.142, respectively.

	For Unplanned CS and Planned CS, the coefficients of Weekend or Holiday are significantly negative. On weekends and holidays, the probability ratio of unplanned CS or planned CS versus VD is 10% to 11% smaller than the ratio on weekdays. For the 3 CS types, the effects of Ghost Month were not significant and the odds ratios are close to 1. Regarding the characteristics of women in labor, the coefficients of Dependent are negative. The factors of a high-risk pregnancy increased the probabilities of the 3 CS types, especially for Planned CS. For women with high-risk pregnancy, the probability ratio of planned CS versus VD is roughly 3.35 times the ratio for women without high-risk characteristics. In addition, the probabilities of Planned CS and Elective CS significantly increased with the age of pregnant women.

	Furthermore, the odds ratios of Period II and Period III were used to test the 2 hypotheses proposed in Rationale, and examine the effects of the policy changes in May 2005 and May 2006. For Unplanned CS, the odds ratios of Period II and Period III are 0.978 and 0.960, respectively. The probability ratio of unplanned CS versus VD decreased only 2% to 4% since May 2005. For Planned CS, the odds ratios of Period II and Period III are 0.995 and 0.942, respectively. Relative to the period before May 2005, the probability ratio of planned CS versus VD did not change significantly between May 2005 and May 2006, but the ratio decreased by approximately 6% since May 2006. Some of the results for Unplanned CS and Planned CS coincide with the 2 hypotheses regarding indicated CS. For Elective CS, the odds ratio of Period II was 0.861, but the ratio of Period III was 1.083. Compared with the period before May 2005, the probability ratio of elective CS versus VD between May 2005 and May 2006 decreased by 14%, but the ratio after May 2006 increased by 8%. The results for Elective CS do not coincide with the 2 hypotheses on elective CS.

	Regarding hospital attributes, the results mean that the CS rates of medical centers, regional hospitals, and local hospitals are all less than the rate of clinics. This is consistent with the results obtained by Xirasagar et al.

 Group practices of hospitals show lower CS rates than individual practice obstetric clinics. Moreover, the public or juristic-person institutes have a higher probability of unplanned CS than private institutes. This is because most private institutes are individual practices, and public or juristic-person hospitals have better emergency departments to handle unscheduled cases.

	The rates of 3 CS types on weekends and holidays were less than those on weekdays, because the timing of planned CS and labor induction could be scheduled in advance. Obstetricians usually scheduled these services on weekdays.

 However, although obstetricians could schedule cases of planned CS and elective CS, a main consideration would have to be the baby’s maturity, such as being at least 37 weeks, and the obstetricians could not advance or postpone the birth date substantially. This could explain the insignificant figures during ghost month.

	From 2003 to 2007, the obstetricians’ legal concerns, time preference, or other convenience motives did not change. In addition, the indications of CS and the monitoring methods adopted by the authorities remained unchanged. The BNHI increased the payment for VD for hospitals by nearly 80% in May 2005, and equalized the provider payment for VD and indicated CS. The results indicate that this supply-side policy slightly lowered the probability of planned CS, but did not reduce the probability of unplanned CS significantly. The policy effect was not great, and the decreasing CS rate was concentrated in planned CS cases. These results imply that the recommendations of unplanned CS cases and most planned CS cases were not affected by this supply-side policy. In other words, the high CS rate in Taiwan could not be attributed totally to the higher payment for indicated CS or to the physicians’ financial incentives.

	Moreover, the BNHI increased the insured’s copayment for elective CS in May 2006. From 2003 to 2007, no other event affected the insured’s attitude toward elective CS. After this demand-side policy change, the rate of elective CS did not decrease. Contrarily, elective CS seemed to increase slightly. This implies that the financial burden might not be an important consideration for women in labor choosing elective CS. Actually, the birth rate in Taiwan has been decreasing yearly. For most families, the medical expenditure of delivery was not a long-run financial burden. The influences of financial burden, such as copayment, might be offset by the joy of giving birth. Therefore, other demand-side considerations, such as preferring an auspicious time or avoiding delivery risk, might be more responsible for the high CS rate in Taiwan.

	Most studies on Taiwan birth data merely discussed supply-side incentives (such as the payment adjustment in May 2005) or focused on demand-side incentives (such as the attitudes of pregnant women).

 Studies discussing the policy effects of the copayment change in May 2006 are scant. This study explores not only the supply side policy in May 2005, but also the demand side policy in May 2006. This is an important difference between this study and relevant literature.

	The analyses for both policies show that the financial incentives were not the main determinant for either the demand side or the supply side in choosing the delivery method (CS or VD). The relevant studies have indicated that fear of delivery would induce the pregnant women’s preference for CS, and obstetricians’ choice of delivery method was influenced by their risk attitude.

 To encourage more VD and lower the CS rate in Taiwan, the authorities should consider mechanisms other than adjusting the payment or changing the copayment. For the demand side, a comfortable environment with enough staff and obstetricians within sight of women in labor would attenuate their anxiety and frustration. For the supply side, providing delivery rooms with enough staff, who are also supportive, and 24-hour emergent facilities could help obstetricians lower the perceived risk of complaints and malpractice litigations. These approaches might help to increase the number of VDs.

	This study does not review all the inpatient records, but adopts the systematic sampling data set as the empirical materials. However, the sampling could represent the population well. The distribution of delivery modes and the characteristics of laboring women in this study are similar to those in the study by Han and Lien, which reviewed all first-birth records between 2003 and 2007.

 Moreover, the analyses of this study lack a control group because the empirical materials were obtained from the NHI research database. NHI in Taiwan is a compulsory program, and more than 99% of the citizens are covered under the NHI system. This study could not observe and discuss the CS rates of women who were not enrolled in NHI and have been absent from these policy changes. Thus, the limitation of this study is that the analysis results might not act as a reference for the delivery modes and incentives of women who are not enrolled in the NHI system.

	This study is based in part on data from the National Health Insurance Research Database provided by the Bureau of National Health Insurance, Department of Health, and managed by National Health Research Institutes. The interpretation and conclusions contained herein do not represent those of the Bureau of National Health Insurance, Department of Health, or National Health Research Institutes.

This study explores whether the policies of financial incentives for the demand and supply sides lowered the Cesarean section rate.

Financial Incentives and Use of Cesarean Delivery: Taiwan Birth Data 2003 to 2007

 Hospital medicine has undergone remarkable growth since its creation. Most (but not all) of the published literature demonstrates better outcomes for patients cared for by hospitalists.

We performed a systematic review and meta-analysis to estimate the magnitude of the impact of hospitalists on length of stay (LOS) and cost.

 Articles published through February 2011 comparing outcomes (LOS and/or cost) of hospitalists with others.

Two reviewers independently searched for abstracted information. We also contacted individual authors to provide us with missing data.

In this systematic review and meta-analysis, we found that hospitalists reduce hospital length of stay without increasing costs.

January 2012

TSI Assay Utilization: Impact on Costs of Graves' Hyperthyroidism Diagnosis

Implementation of Paperless Credentialing in a Multi-State Managed Care Organization

Impact of Cost-Sharing on Treatment Augmentation in Patients With Depression

Financial Incentives and Use of Cesarean Delivery: Taiwan Birth Data 2003 to 2007

The Impact of Hospitalists on Length of Stay and Costs: Systematic Review and Meta-Analysis

Economic Evaluation of Plerixafor for Stem Cell Mobilization

Hypertension Control in Ambulatory Care Patients With Diabetes

Preventive Care and Health Behaviors Among Overweight/Obese Men in HMOs

Systematic Review of Comparative Effectiveness Data for Oncology Orphan Drugs