To evaluate the utility of the Fracture Risk Calculator (FRC, Foundation for Osteoporosis Research and Education) for predicting 10-year hip fracture risk within a “real world” population.

 We identified female members of Kaiser Permanente Northern California aged ≥50 years with bone mineral density (BMD) measured during 1997-2003. Hospitalization for hip fracture was ascertained up to 10 years following the BMD date, and 10-year observed hip fracture probabilities   were calculated. Baseline data for fracture risk calculation were extracted from health plan databases, including age, race/ethnicity, smoking, body mass index, prior fracture, rheumatoid arthritis, glucocorticoid use, disorders associated with bone loss, and femoral neck BMD. Predicted 10-year FRC hip fracture probabilities were compared with observed 10-year hip fracture probabilities.

Among 94,489 women (mean age 62.8 ± 8.6 years, average femoral neck Z-score 0.1), the median duration of follow-up was 6.6 years, during which 1579 (1.7%) hip fractures occurred. Using the FRC, 23% met or exceeded the National Osteoporosis Foundation’s 3% hip fracture threshold. The FRC somewhat underestimated observed hip fracture probabilities; across 10-year risk categories <1%, 1% to 2.9%, and 3% to 4.9%, ratios of observed to median predicted probabilities ranged from 1.3 to 1.4.

The FRC tool can be applied to assess fracture risk in large populations using data from administrative databases. Despite some underestimation, this relatively simple tool may assist targeting of at-risk populations for more complete fracture risk assessment.

	The utility of the Fracture Risk Calculator (FRC) for predicting 10-year hip fracture risk within a “real world” population was evaluated.

		The FRC provided rapid assessment of population fracture risk with some underestimation.

		Bone mineral density as an input parameter appeared to have little overall effect on the tool’s discrimination.

		While the FRC underestimated observed 10-year hip fracture probabilities by 30% to 40%, the ability to rapidly assess fracture risk using population data may be useful for osteoporosis programs in the initial identification of high-risk patients.

	To better comply with current osteoporosis Healthcare Effectiveness Data and Information Set (HEDIS) measures,

 many healthcare delivery systems have developed programs to identify and track women after fracture. Some health plans have also developed quality improvement programs that expand osteoporosis management beyond the HEDIS requirements to include all women and men at increased risk of fracture.

In large part, the stimulus for such targeting has been evidence that preventive osteoporosis treatment for those at increased risk of fracture is cost-effective

 and that comprehensive and widespread interventions can substantially reduce hip fracture rates and thereby produce savings for health plans.

 However, adequate information technology resources must be available to capture information on patients, and management systems must be in place to mount large osteoporosis evaluation and treatment initiatives. Many health plans lack the organization for major fracture prevention outreach efforts, and more importantly, lack a simple, convenient system to identify those at increased risk.

	Several Web-based fracture risk tools are available,

 but these are designed for entry of individual patient-level information and currently do not function in an easily accessible, large-scale, batch mode where data on thousands of individuals can be instantly uploaded for fracture risk estimation. The most widely used fracture risk tool is FRAX, developed by the World Health Organization Collaborating Centre for Metabolic Bone Disease, University of Sheffield, UK

This tool first solicits information on a number of key, independent clinical risk factors and then calculates 10-year probabilities of hip and of any 1 of 4 major osteoporotic fractures. While some of these risk variables are easily obtained from administrative data sets (eg, age, sex), others may be obtained with some effort (eg, race/ethnicity, height and weight, prior fracture, high-dose glucocorticoid exposure, history of rheumatoid arthritis, other conditions known to contribute to osteoporosis, current smoking). Certain variables (eg, family history of osteoporosis, heavy alcohol use) may not be obtained without direct patient questioning. In addition, although many patients have undergone bone mineral  density (BMD) testing, numeric results may not be easily accessible; fortunately, fracture risk tools can produce results with or without BMD. Indeed, fracture risk tools can provide accurate probability estimates with less than complete input data.

 Especially for the purpose of population categorization and targeting osteoporosis outreach programs, results based on a limited data input tool would be considered adequate.

	In this study, we modified an existing Web-based fracture risk tool (the Fracture Risk Calculator [FRC])

 to provide batch outputs using input values obtained from administrative data in a large population of women undergoing BMD testing. We evaluated the performance of the FRC batch tool against observed 10-year hip fracture rates. The FRC tool was chosen because it was highly accessible, efficient, and transparent. Furthermore, the FRC batch estimates were provided at no cost, and results were instantly available via a Web interface. We compared observed 10-year fracture rate estimates with average predicted rates in more than 90,000 female members over age 50 years who received a BMD test in Kaiser Permanente Northern California (KPNC) between 1997 and 2003. Our primary goals were to determine what proportion met or exceeded the National Osteoporosis Foundation (NOF) 3% 10-year hip fracture risk threshold

and to assess whether the tool overestimated or underestimated the true risk of hip fracture.

	Kaiser Permanente Northern California is a large integrated healthcare delivery system serving more than 3 million members in Northern California; approximately 0.5 million are women over the age of 50 years. Since 1991, bone densitometry has been available using Hologic dual-energy x-ray absorptiometry (DXA) scanners (Waltham, MA). By 1995, KPNC had a fully integrated, systemwide patient data collection system that included ambulatory visit diagnoses, radiologic records, and prescription drugs that complemented long-standing hospitalization databases. Therefore, we selected as our study cohort all women aged 50 to 85 years who underwent a hip bone density scan on a Hologic scanner (models QDR 2000, 4500, or Delphi), selecting the first scan during 1997-2003. We excluded those who did not have at least 1 year of continuous (<90-day gap) membership both prior to and following the DXA scan date, those for whom DXA data were not electronically accessible, and those with missing race/ethnicity. We also excluded women who had filled a prescription for a bisphosphonate in the year prior to the DXA test. The study was approved by the Institutional Review Board of KPNC.

	The fracture model estimate began with population-based 10-year fracture probability for age, sex, and race/ethnicity. Next, specific patient characteristics were compared with those of the base population, and relative risks were applied to factors that differed between the individual patient and the base population. In very simplistic terms, the product of base rate times the risk differences yielded the predicted absolute 10-year risk. The current base US 10-year fracture risks for men and for women used by the FRC are those calculated from the 2006 US National Inpatient Survey by Ettinger et al.

 The FRC model’s relative risks for various clinical risk factors are shown in 

. A detailed manual of operations for the Web-based FRC batch tool is available through the Foundation for Research and Education (www.fore.org).

	Age, race/ethnicity, and body mass index (BMI) were determined at the index BMD scan date. Those with missing BMI were assigned a null value of 25 kg/m2, which was the median value in our cohort. Using ambulatory care, hospitalization, and pharmacy databases, we obtained each patient’s data from the 1 year prior to the DXA measurement to secure the following exposures and diagnoses: glucocorticoid use ≥1825 mg of cumulative prednisone dose equivalent in the prior year (average 5 mg/day), rheumatoid arthritis diagnoses, and secondary causes of bone loss (diabetes mellitus with insulin use, malabsorption syndrome, chronic liver disease, and osteogenesis imperfecta). We determined prior history of fracture after age 45 years (up to 10 years prior to the DXA date) based on hospitalization and outpatient diagnoses of fracture (

International Classification of Diseases,Ninth Revision

] codes 800-829), excluding open fractures, fractures related to severe trauma (

 E-codes 800-845), and fractures of the fingers, toes, facial bones, and skull, since these are not generally considered to be osteoporotic. Information on alcohol consumption and parental history of hip fracture was not available, and smoking status was not uniformly available. We assumed that all missing input values were null.

	Under the auspices of institutional agreements for protection of proprietary interests, batch data without patient identifiers were securely uploaded to the Foundation for  Osteoporosis Research and Education FRC Web site. Ten-year fracture probabilities were returned, calculated both with and without BMD inputs.

Follow-up and Observed Hip Fracture Outcome

	Individual patient data were examined from the index BMD scan date until the earliest of the following: the fourth prescription for a bisphosphonate, likely indicating 1 year of exposure because typical prescriptions provide a 3-month supply (N = 19,440, 20.6%); a principal diagnosis of hip fracture (

 code 820.0X, 820.2X, and 820.8X, excluding open fractures and those associated with major trauma, 

 E-codes 800- 845); death; disenrollment (>90-day gap in membership); or when 10 years had elapsed from the index date.

	Differences between women with and without subsequent hip fracture were compared using the 

2 test for categorical variables and the Student t test for continuous variables. Incident hip fracture rates and 95% confidence intervals (CIs) were calculated per 1000 person-years up to 10 years followup, the censoring date, or December 2009 (whichever occurred first). We used Kaplan Meier product-limit estimates to calculate the 10-year hip fracture probabilities from observed events over time and compared these with the median predicted 10- year hip fracture risk for the subcategories <1%, 1% to 2.9%, and 3% to 4.9%, as well as 5% to 6.9%, 7% to 9.9%, and >10%. We also compared the proportions of women in each decade of age meeting or exceeding the NOF’s 3% cost-effectiveness hip fracture risk threshold

both with and without inclusion of BMD as an input parameter. We used the area under the receiver operating characteristic (ROC) curve (C statistic) to compare sensitivity and specificity of the FRC results, with and without BMD.

 The C statistic ranges from 0.5 (the predictions are no better than chance) to 1.0 (a perfect predictive model). Statistical analyses were performed using SAS version 9.1 (SAS Institute, Cary, NC) and Stata version 9.2 (Stata- Corp, College Station, TX).

	Hologic BMD data were available for 94,489 women between the ages of 50 and 85 years, after excluding those not meeting health plan membership criteria (n = 19,178), those with missing race/ethnicity (n = 318), those with missing BMD data (n = 257), and those who received bisphosphonate drugs in the year prior to the index BMD date(n  = 2730).

shows the characteristics of the cohort at baseline. The mean femoral neck Z-score was 0.14, suggesting that the study population overall was comparable to the Hologic reference range (National Health and Nutrition Examination Survey III).14 The median duration of follow-up was 6.6 years (interquartile range 3.6-8.3 years), during which 1579 hip fractures were observed. Those who subsequently suffered a hip fracture were older, more likely to be white, had lower BMI, had other risk factors for osteoporosis, and were less likely to be users of hormone therapy. The mean age at hip fracture was 78 years and hip fracture incidence rates increased markedly with age (

). The incidence of hip fracture in nonwhites (Asians, Blacks, and Hispanics) was approximately half the incidence in whites.

	When applying the NOF’s 3% 10-year hip fracture probability threshold, the FRC yielded an overall population proportion of 23% with a strong relationship to age (

). Thus, the proportion of women qualifying under this criterion for osteoporosis management increased from about 1 in 25 at age 50 to 59 years to about 4 in 5 women at 75 years and older. The overall proportion identified as high risk showed no change when BMD was removed from the calculator input; however, when BMD was excluded, the proportion was lower among women aged 60 to 69 years (7.2% vs 17.1%) and higher among women aged 70 to 79 years (81.1% vs 59.4%).

 shows the predicted 10-year hip fracture probabilities compared with the observed 10-year hip fracture probabilities. The FRC underestimated the true fracture probabilities by 30% to 40%, particularly in the higher predicted risk categories (

). Although 40% of our cohort used hormone therapy at baseline, a factor not included in the FRC model, this did not explain the model’s tendency to underestimate the observed fracture probability (data not shown). The area under the ROC performances of the FRC tool were similar regardless of whether BMD was included (C statistic = 0.85; 95% CI 0.84, 0.86) or not (C statistic = 0.83; 95% CI 0.82, 0.84).

	We found that the FRC, a relatively simple, transparent fracture risk tool, provided rapid assessment of population fracture risk with some underestimation. When applied to our large population of women 50 years and older, approximately 1 in 4 were found to potentially qualify for osteoporosis treatment based on the NOF threshold of 3% 10-year hip fracture risk. Although the FRC somewhat underestimated the observed fracture probabilities in our cohort, similar underestimates have been reported by Sornay-Rendu et al when using FRAX to assess the risk of major osteoporotic fracture in a cohort of French women.

 Both the FRC and FRAX use the same base rate for 10-year fracture risk in women as this rate is provided by the same source

; the same risk input variables are also used, although the relative risks differ

 The FRAX model is also more complex mathematically, because it uses Poisson regressions to accommodate interaction terms and to calculate fracture probability offset by expected mortality.

	The inclusion or exclusion of BMD as an input parameter for the FRC batch calculator appeared to have little effect overall on discrimination, as measured by the C statistic (C statistic 0.85 and 0.83 with and without BMD, respectively). Several studies have also shown that FRAX yields similar C statistic estimates in the range of 0.7 to 0.8,

 including comparison between different models or with different input variables.

 It should be recognized, however, that there may be limitations to the utility of ROC curves in assessing clinical risk prediction since very large relative risks of included (or removed) variables are necessary to change the statistical outcome.

	We believe that both the calibration and discrimination of the FRC tool would be improved with more complete data input, particularly with inclusion of family history datain  the model. Epidemiologic studies indicate that parental history of hip fracture increases from about 5% in women in their 50s to about 15% to 20% for women 70 years and older

; the model’s fracture prediction would nearly double for women with such a history. In addition, our ascertainment of prior fractures, limited to 10 years prior to baseline, yielded a prior fracture prevalence of about half that reported in epidemiologic studies

; here too, the model predictions would nearly double for women with such a history. Other missing data (eg, alcohol intake missing for all, BMI missing for 30% of the cohort) are less important in the model’s calibration since thinness and heavy alcohol intake are not common in aging women.

These factors, along with more complete ascertainment of smoking status and other conditions contributing to bone loss, could be important when calculating risk for an individual.

	Our analyses did not examine thresholds for clinical actions and further did not consider cost-effectiveness of such actions. The threshold for treatment depends on available resources; Kanis et al have suggested for the United Kingdom a changing threshold dependent on age,

 while in the United States, the NOF thresholds of 3% hip fracture and 20% major osteoporotic fracture probabilities are applied to women irrespective of age.

 Recently, the US Preventive Services Task Force has suggested thresholds for densitometry screening based on 10-year FRAX fracture risk thresholds of 1.2% for hip fracture and 9.3% for any 1 of 4 major osteoporotic fractures (probabilities equal to the 10-year risks for a 65-year-old white woman without risk factors).

 We note that FRC is reasonably accurate within this risk range.

	The strengths of our study include the large cohort size with more than 1500 hip fracture outcomes and the fact that comprehensive data on both risk factors and outcome were available in the context of an integrated health plan system. Given the high membership retention rates in this population, we had a long follow-up period for hip fracture, accruing more than 570,000 patient-years.

	Our study had the following limitations. First, our study population included only women undergoing BMD testing and thus may have included a somewhat higher proportion of younger midlife women. However, our cohort had 34,529 women older than 65 years, and the mean Z-scores of our population suggest our cohort was similar to the National Health and Nutrition Examination Survey III reference populations used by Hologic, Inc.

 A second limitation was missing or incomplete data for the fracture tool  risk factors (discussed above). Others have also shown that less-than-complete risk data input appears to detract only a little from fracture risk prediction models,

 and a comprehensive assessment of clinical risk factors would be part of the appropriate next step prior to making a clinical treatment decision. Third, we ascertained only hip fracture outcomes and cannot comment on the tool’s ability to predict any 1 of 4 major osteoporotic fractures in our population. Determining incident spine fractures from administrative data is extremely difficult without individual chart review. Given the close relationship between hip and any 1 of 4 major osteoporotic fractures,

 we predicted  similar performance of the tool for this outcome. In fact, Leslie and colleagues have shown that one can accurately impute major osteoporotic fracture rates from hip fracture rates.

 Finally, we did not have complete 10-year follow-up on all patients and thus the observed 10-year fracture rates were estimated.

	A major barrier to use of any new clinical tool is its acceptance by providers and support by their institutions. The tool should also be easy to use, require little effort to run, and provide data to patients and providers that are easily grasped and interpreted. After initial targeting of at-risk patients, fracture probability should be assessed at the individual level, whereby decisions regarding aggressive osteoporosis management can be made once patients are properly evaluated and counseled. The FRC batch tool is meant to be used as a population stratification tool and is not intended to replace individual risk assessment. This approach may be helpful not only for systemwide quality improvement efforts but also for assessing potential thresholds for large-scale outreach efforts based on available resources.

	In summary, it is now possible to apply a relatively simple risk tool using secure Web-based uploading of batched data in order to acquire aggregate fracture risk  information. This population information may be useful for resource planning and integration into osteoporosis quality improvement programs to assist in rapid initial identification of those at risk for fracture.

	These data were presented in part at the 92nd Annual Meeting of the Endocrine Society; June 19-22, 2010; San Diego, CA. The authors would like to thank Ryan Navarro, BA, and Mohammad Hararah, BA, for their assistance in assembling the bone mineral density data.

Articles

CLINICAL

The FORE Fracture Risk Calculator provided rapid assessment of population hip fracture risk with some underestimation compared with observed hip fracture probabilities.

Fracture Risk Tool Validation in an Integrated Healthcare Delivery System

To assess the level of anticoagulation control achieved in patients with atrial fibrillation (AF) receiving routine medical care within a large managed care organization and to explore patient factors that influence control.

Retrospective cross-sectional study of all patients with AF treated in Clalit Health Services (CHS) community clinics in central Israel between November 1, 2006, to October 31, 2007.

Using the CHS computerized database, we identified 906 patients with a diagnosis of AF who were treated with warfarin for at least 6 months. Data included patient demographics, comorbidities, and international normalized ratio (INR) values as well as managing physician certification. Anticoagulation control was assessed by measurement of time within therapeutic range (TTR) (INR 2-3). Univariate and multivariate analyses were performed to explore the association of patient variables with anticoagulation control.

 Roughly two-thirds of patients had poor anticoagulation control, as evidenced by TTR of <60%; the mean TTR was 48.6%. Poor control was significantly associated with female sex, advancing age, and comorbid conditions. Heart failure and having a non—board-certified physician were found to be independent predictors of poor control (odds ratio [OR] = 1.63; 95% confidence interval [CI] = 1.20-2.22; and OR = 1.41; 95% CI, 1.05-1.88, respectively).

 Quality of anticoagulation in patients with AF receiving routine medical care was suboptimal, with nearly half the time spent outside the therapeutic range. Ways to improve anticoagulation control among patients with AF should be sought.

	Because adequate anticoagulation control in patients with atrial fibrillation is of medical and economical importance, it should be optimized.

		In the present study, patients with atrial fibrillation receiving routine medical care within a large managed care organization had suboptimal anticoagulation control.

		Patients receiving oral anticoagulants spent less than half the time within the recommended therapeutic range, and therefore were at heightened risk for medical complications, mainly stroke.

		Patients with comorbidities were more likely to have poor anticoagulation control.

Atrial fibrillation (AF), the most common cardiac rhythm disorder, heightens the risk for ischemic stroke 4- to 5-fold.

 The use of oral anticoagulants such as warfarin has been shown in clinical trials to reduce the risk of stroke by 64%; thus, warfarin therapy is widely accepted in patients with AF and is advocated by the American College of Chest Physicians.

In order to achieve maximal protection against stroke and to minimize bleeding complications, warfarin therapy must be tightly controlled and maintained within a narrow therapeutic index of international normalized ratio (INR) values between 2 and 3. This task is by no means trivial as each INR determination, which requires a venipuncture, needs to be promptly addressed by the managing physician. Moreover, INR levels are influenced by an array of factors including patient age, comorbidities, concurrent medications, genetic makeup, and diet.

 As a result, oral anticoagulant therapy necessitates regular and diligent monitoring, which can be toilsome for patients and physicians alike.

Although not easily achieved, high anticoagulation control, expressed as the time spent within the therapeutic range (TTR), has a paramount affect on patient outcomes, reducing stroke events and mortality rates.

 Moreover, it is estimated that optimal anticoagulation could prevent 28,000 cases of stroke in the United States annually, leading to a $2.5 billion cost reduction.

Even though the literature acknowledged the superior outcomes of anticoagulation clinics over routine medical care in terms of anticoagulation control, anticoagulation management often is in the primary care physician’s domain.

 Nevertheless, there is a relative paucity of data concerning the quality of anticoagulation achieved in routine medical care, although it is assumed to be the most prevalent form of anticoagulation care in the United States.

 Moreover, studies that addressed anticoagulation care in the community were seldom population based; thus, they had selection bias that limited their generalization to other populations.

Also, previous studies looking at anticoagulation control in the managed care setting had a heterogeneous patient population (ie, some patients received care in the community, while others were treated in anticoagulation clinics), which interfered with evaluation of the anticoagulation control achieved in routine medical care.

In this article we describe the quality of anticoagulation control achieved in patients with AF receiving routine medical care within a large managed care organization (MCO) in Israel. The purpose of this study was to assess the quality of anticoagulation control (expressed as TTR) and to explore patientlevel factors that may have affected it.

This study was carried out in the central district of Clalit Health Services (CHS), Israel’s largest government-funded MCO. The central district of CHS provides medical care to approximately 500,000 patients residing in central Israel, a largely urban setting. All patients had full medical coverage by CHS inclusive of pharmacy benefits for prescription medication as granted to all Israeli citizens by order of the National Health Insurance Act.

Following approval of the CHS local institutional review board, we conducted a retrospective study from November 1, 2006, to October 31, 2007, using the CHS computerized database to identify all patients with a diagnosis of AF who were treated with warfarin for at least 6 months. Patients were excluded if they fulfilled any 1 of the  following criteria: (1) were younger than 18 or older than 85 years; (2) were elderly and lived permanently in a nursing home; (3) had an active malignancy; (4) had prosthetic heart valves; (5) were bedridden; (6) were prescribed antipsychotic medication; or (7) had fewer than 5 INR determinations during the study period. All records retrieved from the database were audited manually by study staff for concordance with the above-mentioned criteria.

A total of 906 patients met the study criteria and were included in the analysis. Each patient was managed by his/ her personal physician during the study period. Overall, care was delivered by 124 primary care physicians in CHS community clinics. The computerized database provided demographics (age, sex) and medical diagnoses. In addition, the number and value of INR determinations for each patient were also extracted. Data on physicians’ board certification were retrieved from administrative records.

Anticoagulation control was assessed by measurement of time spent within the TTR (ie, time in which patient INR values were between 2 and 3). The therapeutic range was calculated with computer software that utilized a linear interpolation model, as described by Rosendaal et al.

 First, the TTR was determined for each patient. Later, stratification of patients according to TTR level was carried out as follows: a TTR level <60% was considered to represent poor anticoagulation control, a TTR level between 60% and 75% was considered to represent good anticoagulation control, and a TTR level >75% was considered to represent excellent anticoagulation control. This stratification allowed characterization of patient subsets associated with the different control levels.

All statistical analyses were performed using SPSS, version 15.0 (SPSS Inc, Chicago, IL). Each potential predictor of poor control was first assessed in univariate models (

² test for categorical variables and analysis of variance for continuous variables). Significant univariate predictors were subsequently assessed in the multivariate logistic regression model to determine their independent effect, expressed as odds ratio (OR) and 95% confidence interval (CI). P <.05 was considered significant.

A total of 906 patients with AF who were treated with warfarin for at least 6 months were identified through the computerized database. 

 presents patient demographics and clinical characteristics. The mean age was 71.7 years; 51.9% were female and more than 90% of patients had at least 1 risk factor for ischemic stroke (age >75 years, diabetes mellitus, hypertension, heart failure, or prior stroke). Patients were receiving routine medical care delivered mainly by non—board-certified physicians and by board-certified family physicians (48.6% and 37.1%, respectively).

Patients had 769 patient-years of follow-up (mean 310.6 days per patient), during which 14,935 INR determinations were performed. Due to the interpolation method, 137 patient-years could not be evaluated for TTR since INR determinations were performed more than 30 days apart.17 Patients had a mean of 16.5 INR determinations during the study period (range 5-75) and spent 48.6% of the time within the therapeutic range of 2 to 3, 32% of the time under the therapeutic range, and 19.3% of time above the therapeutic range (

When patients were stratified according to anticoagulation control levels (TTR <60%, TTR 60%-75%, TTR >75%), more than two-thirds of them had poor anticoagulation control (

). Only 11.9% had excellent anticoagulation control, and 20.6% had good anticoagulation control. Compared with the group that had poor anticoagulation control, the group that had excellent anticoagulation control had younger patients and fewer females (P = .006 and P = .02, respectively). Additionally, poor anticoagulation control was associated with more frequent INR testing than excellent control. It was also noticeable that the excellent-control group was less burdened by the comorbiditiesof diabetes, heart failure, and stroke (P = .003, P <.001, and P = .001, respectively).

Patients with poor anticoagulation control were seen more often by non—board-certified physicians than patients with excellent anticoagulation control (53% poor controlvs   40% excellent control, P = .018). An opposite trend appeared among board-certified family physicians, but it did not reach significance (36% poor control vs 45% excellentcontrol, P = .096) (Table 3).

In order to evaluate the independent effect of each variable as a predictor of poor anticoagulation control, we performed a multivariate logistic regression (

). We identified 2 significant predictors of poor anticoagulation control: having a non—board-certified physician and heart failure (OR = 1.41; 95% CI, 1.05-1.88; and OR = 1.63; 95% CI, 1.20-2.22, respectively).

In our study, patients with AF receiving routine medical care within a large MCO had suboptimal anticoagulation control with a mean TTR of 48.6%. Additionally, poor anticoagulation control was associated with comorbidities and having a non—board-certified physician. Since a close correlate between anticoagulation control and clinical outcomes (ie, stroke, bleeding events) exists, suboptimal control has profound medical and economic implications.

Our results are fairly consistent with those of previous studies where anticoagulation control in routine medical care was assessed, as evident in a recent meta-analysis by Baker et al that found community-based AF anticoagulation control to be 51% (95% CI, 47%-55%) and anticoagulation clinic control to be 63% (95% CI, 58%-68%).

 This explains to some extent the lower efficacy of oral anticoagulants in stroke prevention in the community setting.

 Different patient characteristics are a possible reason for discrepant control levels in clinical trials/anticoagulation clinics compared with community clinics.

This may be especially true among populations burdened by comorbidities such as ours. Indeed, our patient population had a heightened prevalence of comorbidities (diabetes 28.7%, heart failure 41.0%, and stroke 18.3%). All of these comorbidities were significantly associated with poor TTR control (P = .003, P <.001, and P = .001, respectively). It is noteworthy that heart failure has been the only medical comorbidity found to be an independent predictor of poor TTR control (OR = 1.63; 95% CI, 1.20-2.22). An explanatory mechanism is the possible interaction between warfarin and multiple drugs administered to heart failure patients.

Additionally, our data imply that older age is more prevalent in the poor control group (P = .006), unlike results reported by Rose et al, which did not show an age difference  among anticoagulation control groups.

 It is plausible that this finding is a surrogate for the higher burden of comorbidities associated with increasing age. Also, our study found a greater proportion of women in the poorly controlled group (P = .021), consistent with the results of the aforementioned study. Further research is needed to explore sex-related differences in anticoagulation control, especially in light of the heightened risk for stroke among women with AF.

As mentioned above, anticoagulation control achieved in anticoagulation clinics and in clinical trials is superior to that achieved in community clinics.

 Nevertheless, control levels attained in different community settings vary widely. While some studies report high-quality anticoagulation control with TTR levels above 65%,

 others demonstrate poor anti- coagulation control with TTR levels below 50%.

 Aside from patient selection bias that could explain the above-mentioned discrepancy, our study suggests that physicians’ certification also had an effect on anticoagulation control even within the same setting. Hence, more patients managed by non—board-certified physicians as opposed to patients seen by board-certified family physicians were found in the poor control group. Indeed, having a non–board-certified physician was an independent predictor of poor anticoagulation control (OR = 1.41; 95% CI, 1.05-1.88).

This finding may be related to physician attitudes and knowledge concerning anticoagulation care. With regard to the delicate balance of benefits versus risks in oral anticoagulant therapy, some physicians tend to undertreat patients because they fear a bleeding complication, even at the expense of failing to prevent an ischemic stroke.

 Also, lack of clear practice guidelines regarding optimal scheduling of INR tests may hinder delivery of optimal anticoagulation care by physicians, as described by Shalev et al.

 In fact, significantly more patients in the poor-control group had an above-average number of INR determinations (43.5 % vs 25.9%, P = .001), which may indicate precariousness in anticoagulation care. This issue needs to be further explored, but it may relate to lack of adequate training in anticoagulation care among non—board-certified physicians.

Unfortunately, in our study, patients on warfarin therapy were within the recommended therapeutic range less than half the time. A number of studies both in the United States and in Europe evaluated the economic benefit associated with optimization of anticoagulation control among patients with AF.

 Substantial cost savings stemmed mainly from stroke prevention but also from reduced hospitalization rates and emergency department visits.

Since optimal anticoagulation control is desirable on both medical and economic grounds, ways to improve control should be sought. If good anticoagulation control cannot be achieved within the usual care setting, specialized anticoagulation clinics are a validated alternative option.

 Moreover, innovative methods are being examined, some with promising results, such as handheld patient INR meters and computer- aided programs for warfarin maintenance.

Despite the above-mentioned efforts, warfarin may not ultimately provide the optimal anticoagulation needed. Therefore, its substitution with newer oral anticoagulant drugs may eventually be inevitable. The direct thrombin inhibitor dabigatran, which abolishes the need for INR monitoring, has recently proved its efficacy and may be the  anticipated substitute for warfarin.

 As the newer anticoagulant drugs are associated with substantial expenditures, a careful cost—benefit analysis should be conducted to determine their feasibility.

Our study has several limitations. First, we were unable to acquire data concerning clinical outcomes such as stroke and bleeding event rates for our study population. For that reason, we elected to use anticoagulation control as a surrogate indicator for outcome, given the strong association between TTR levels and clinical outcomes.

 Second, we could not assess scheduled interruptions of oral anticoagulants (ie, periprocedural, hospitalization), which may have resulted in underestimation of the TTR levels of the study group; however, a similar study estimated interruptions to cause a 5.6% decline in TTR levels, which does not alter the results considerably.

 Finally, our population had a high burden of comorbidities, which may limit the study’s generalizability to other settings, although anticoagulation control achieved in our study is similar to that found in a meta-analysis examining anticoagulation quality in patients with AF reported by Baker et al.

This study provides important information about anticoagulation control of patients with AF who receive routine medical care within a large MCO. Overall, patients with AF had suboptimal control, with less than half the time spent within the therapeutic range, which placed them at heightened risk for medical complications, mainly stroke. Our results suggest that patient comorbidities and lack of physician board certification negatively affect anticoagulation control. Patients with AF who are treated in routine care could benefit from methods aimed at improving control; hence, further research is needed to assess cost-effectiveness of such methods.

	The authors wish to thank Professor Frits R. Rosendaal, MD, PhD, Leiden University Medical Center, The Netherlands, for kindly providing the computer software used in data analysis.

 From the Department of Family Medicine (OCM, GH, AE, SV), Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel; Department of Family Medicine (AE, SV), Central District, Clalit Health Services, Rishon LeZion, Israel; and Department of Family Medicine (OCM), Maccabi Healthcare Services, Tel Aviv, Israel.

MANAGERIAL

Patients with atrial fibrillation receiving routine medical care within a large managed care organization were found to have suboptimal anticoagulation control.

Quality of Anticoagulation Control Among Patients With Atrial Fibrillation

To investigate the effects of highdeductible health plans (HDHPs) and consumerdirected health plans (CDHPs) on healthcare spending and on the use of recommended preventive care.

 We analyzed claims and enrollment data for 808,707 households from 53 large US employers, 28 of which offered HDHPs or CDHPs. We estimated the effects of HDHP or CDHP enrollment on healthcare cost growth between 2004 and 2005 using a difference-in-difference method that compared cost growth for families who were enrolled in HDHPs or CDHPs for the first time in 2005 with cost growth for families who were not offered HDHPs or CDHPs. Control families were weighted using propensity score weights to match the treatment families. Using similar methods, we examined the effects of HDHP or CDHP enrollment on the use of preventive care and the effects of HDHP or CDHP offering by employers on the mean cost growth.

 Families enrolling in HDHPs or CDHPs for the first time spent 14% less than similar families enrolled in conventional plans. Families in firms offering an HDHP or a CDHP spent less than those in other firms. Significant savings for enrollees were realized only for plans with deductibles of at least $1000, and savings decreased with generous employer contributions to healthcare accounts. Enrollment in HDHPs or CDHPs was also associated with moderate reductions in the use of preventive care.

The HDHPs or CDHPs with at least a $1000 deductible significantly reduced healthcare spending, but they also reduced the use of preventive care in the first year. This merits additional study because of concerns about enrollee health.

	The effect of enrollment in high-deductible health plans (HDHPs) or consumer-directed health plans (CDHPs) on healthcare spending and on the use of preventive care was assessed across multiple employers, insurance carriers, and plans in a 2-year retrospective study.

		Families enrolling in HDHPs or CDHPs and firms offering HDHPs or CDHPs spent less on healthcare.

		Significant savings were realized only for enrollees in plans with deductibles of at least $1000, and savings decreased with generous employer contributions to healthcare accounts.

		Enrollment in HDHPs or CDHPs was associated with moderate reductions in the use of preventive care, despite the fact that these plans waived the deductible for preventive care.

Curbing increases in healthcare costs is a top priority for policy makers and for employers. Many believe that high-deductible health plans (HDHPs), also known as consumer-directed health plans (CDHPs), when coupled with personal savings accounts, might be one way to hold down costs. These plans, intended to make patients more cost conscious, are becoming increasingly popular, and healthcare reform may foster further growth in enrollment. As of 2009, 20% of Americans with employer  coverage were enrolled in a plan with a deductible high enough to be eligible for a health savings account. Among those purchasing coverage directly, 47% had a deductible at least this high.

 A survey of large employers at the beginning of 2010 found that more than 54% offered at least 1 CDHP option, and another 7% were planning in 2011 to adopt one.

 Growth is expected from 2 sources, namely, CDHPs with low premiums offered through health insurance exchanges and more CDHP offerings in the employer market because of taxes placed on generous “Cadillac” plans.

Despite growing enrollment, little is known about the effects of HDHPs or CDHPs on healthcare costs and on the use of necessary care. Even less is known about the influence of specific HDHP or CDHP provisions, including deductible levels and account offerings.

 Both questions are of key importance for those who are newly insured  through exchanges and for those who are selecting plans in the employer or individual market. Part of the problem is the lack of good pre-post data for persons enrolling in diverse HDHP or CDHP and conventional plans. Most evidence is limited to studies with data from a single carrier, a single employer, or a single year; therefore, the findings may not apply outside of those settings. A review of these studies concluded that moving consumers from traditional plans to high-deductible plans could result in significant savings; however, coupling these plans with funded personal accounts could reduce this effect.

 More recent work suggests that some CDHP plan designs might lead to higher spending over time, to discontinuation of chronic disease medications by patients, and to decreased use of office visits, hospitalizations, and emergency department care.

 have found instances in which CDHP enrollment has no discernible effect on the use of preventive care. Reviewing a set of industry studies, the American Academy of Actuaries

 concluded that CDHP plans saved 12% to 20% in their first year compared with control plans, with no evidence that this was due to a reduction in necessary care.

This study is the first to date to use longitudinal data from a diverse set of carriers, employers, and HDHP or CDHP designs to investigate the first year effects on healthcare spending and on the use of preventive care. It also is the first study to examine the differential effects on these outcome measures of deductible levels, personal accounts, and employer account contributions.

We constructed a unique data set that included 2 years of enrollment and healthcare claims information for employees of 53 large US employers, 28 of which offered HDHPs or CDHPs to their employees. These employers offered health plans from all major and many smaller US insurance carriers. We defined a high deductible as $500 or more for single coverage and $1000 or more for family coverage in 2005, resulting in plans with a range of deductible levels and account provisions.

We estimated the differences between HDHP or CDHP and non—HDHP or CDHP enrollees using a difference-indifference propensity score–weighted method. We compared the 2004 to 2005 change in healthcare costs for families who first enrolled in HDHPs or CDHPs in 2005 (treatment families) with the change in healthcare costs for families who remained in conventional plans (control families). Therefore, the analysis controls for all time-invariant differences across treatment and control families such as inherent propensity to use healthcare or trust in physicians and modern healthcare. However, our results might be confounded if treatment and control families had different cost growth trajectories. To address this, we took 3 additional steps. First, we only included families as controls if they were not offered an HDHP or a CDHP. Those who were offered and declined an HDHP or a CDHP were excluded from the analysis. Second, we used propensity score weights to produce a control group similar to the treatment group based on a rich set of observed characteristics, including family type, geocoded income and educational levels, presence of major diagnoses, actuarial value of the health plan before enrollment, and industry of employment.

 Third, to account for residual confounding, we used these observed characteristics as covariates in multivariate regression models.

We also conducted an intent-to-treat analysis to estimate the effects on healthcare cost growth of the employer decision to offer HDHPs or CDHPs. Using difference-in-difference methods, we compared the 2004 to 2005 healthcare cost increases for employers who offered HDHPs or CDHPs in 2005 versus for employers who did not. We also estimated separate effects among employers who completely replaced their conventional plans with HDHPs or CDHPs.

The study population consisted of active full-time employees and their dependents who were continuously enrolled for 2 full plan-years. A small proportion (0.4%) were dropped because of errors or omissions in their claims data. A slightly larger proportion (3.1%) were dropped because of errors in their enrollment information. This resulted in 808,707 families for analysis related to the effects of HDHP or CDHP enrollment and 981,973 families for analysis related to the effects of HDHP or CDHP offer.

The employers entered the study from 2 routes. One group of employers was recruited because they offered an HDHP or a CDHP during the period from 2003 to 2007. These employers were selected to encompass a range of geographic regions, employee income levels, proportion of employees enrolling in HDHPs or CDHPs, and HDHP or CDHP characteristics. The other group of employers was from the Thomson Reuters (New York, New York) MarketScan database. These employers were selected to match the geographic, size, and industry distribution of the recruited employers. In the 2004-2005 cohort used for this analysis, employers from both sources  contributed to the treatment and control samples (83% of HDHP or CDHP—enrolled families are from recruited firms). The enrollment and claims data from insurers were  standardized into a modified MarketScan format. An expert independent of the study organizations certified that the analysis data files received by the research team were deidentified, and the Human Subjects Protection Committee at RAND Corporation approved the study.

Families are the unit of analysis, with additional variables indicating a single employee, employee plus spouse, and additional tiers. For the effect of HDHP or CDHP enrollment analysis, treatment families were those who first enrolled in an HDHP or a CDHP in 2005. For the effect of HDHP or CDHP offer analysis, the treatment families included all insured families in firms that first offered an HDHP or a CDHP in 2005. In both cases, the treatment group was restricted to those who worked for employers where at least 3% of employees enrolled in an HDHP or a CDHP. Control families worked for employers that did not offer high-deductible plans.

High-deductible health plans are classified into the following 4 types by individual deductible and by employer contribution to personal medical accounts: (1) moderate deductible ($500-$999), (2) high deductible (>$1000) with no account, (3) high deductible with low employer account contribution of less than $500, and (4) high deductible  with generous employer account contribution of at least $500 (the last 2 are also known as CDHPs); the types represented 44%, 11%, 33%, and 13% of the treatment sample, respectively. Almost all of these high-deductible plans waived the deductible for preventive care, as established by employer survey and interview data.

We derived plan cost-sharing provisions for all plans based on payment patterns in the claims data combined with employer survey data if available. We included in our analysis only plans with at least 100 employees to ensure sufficient observations to make reliable estimates of the deductible, which is used to assign treatment status. We validated our claimsbased cost-sharing provisions by comparing them with survey responses from 27 employers about 138 plans they offer with a total enrollment of 1.1 million members in 2005. Comparing the treatment classification based on the 2 sources, we found agreement for 93% of enrollees. In addition, all high-deductible plans identified for this analysis were confirmed by survey data or other communication with the employer.

We calculated annual family costs for medical care (insurance and patient payments for care received) and divided these by 12 to obtain the mean monthly expenditures. Parallel calculations resulted in the mean monthly expenditures in each of the following 4 healthcare settings: outpatient, inpatient, emergency department, and prescription drugs.

The following 6 preventive care outcomes were created based on Healthcare Effectiveness Data and Information Set (HEDIS) measure definitions

measures, receipt of mammography, cervical and colorectal cancer screening, and glycosylated hemoglobin (A1C) testing for patients with diabetes mellitus. Dichotomous  measures were created at the annual family level indicating whether some or none of the eligible family members had obtained the recommended care. We adapted HEDIS 2008 specifications to conform to the single-year windows in our analysis framework (discussed further in the 

). We created 2 child immunization measures indicating whether a child was on track to obtain the full set of recommended immunizations. Counting only care received in the current year caused these measures to be lower than typical HEDIS measures but consistently so in each year and across the treatment and control groups.

Covariates for analyses were derived from enrollment, claims, and geocoded location. Enrollment files provided family type, age of head of household, family size,  geographic region, metropolitan statistical area status, and employer’s industry type. Claims data supplied prospective relative risk scores based on diagnostic cost group

summed to the family level and indicators for whether a family received care in each of 23 major diagnostic categories. Actuarial values (percentage of allowed charges paid by the plan) were calculated for each plan using the plan provisions to simulate payment of claims for a standard population. The zip code—level geocoded characteristics are the median household income, percentage of adults with high school and college degrees, percentage unemployed, and percentage of Hispanic, black, and non- Hispanic white race/ethnicity.

For the cost-outcome models of the effects of enrollment in HDHPs or CDHPs, we used propensity score weighting to balance the distributions of numerous characteristics observed in 2004 between treatment and control families.

 Logistic regression analysis was used to model the odds of being a treatment family as a function of characteristics that predict both health plan selection and healthcare use (discussed further in the eAppendices). Predicted probabilities (propensity scores) were used to derive individual family weights for control families proportional to the conditional probability of being a treatment family. To check the adequacy of the propensity score  model, we evaluated the balance of the weighted means of the measured characteristics. When balance is obtained, weighted analyses adjust for potential confounding owing to measured characteristics.

When estimating the effects for different types of high-deductible plans, treatment families were divided into different subgroups. Each subgroup was propensity score weighted to match the distribution of characteristics of the entire treatment sample (discussed further in the eAppendices).

Most of the covariates included in the propensity score model were also included in the weighted outcome models to provide estimates that are more efficient and “doubly robust” to misspecification of either model.

 We tested a range of generalized linear model specifications (identity and log links and constant, proportional to the mean, and proportional to the mean squared variance functions) to address the skewness and truncation at zero in healthcare costs.

 None of the other models tested outperformed the identity link and constant variance; hence, this is the model specification we use. Robust standard errors that account for clustering of family over time were used in all models. To estimate the effects of the employer decision to offer HDHPs or CDHPs on healthcare cost growth, we used parallel models but without propensity score weighting of those families not offered HDHPs or CDHPs.

For the HEDIS immunization outcomes, we computed unadjusted difference-in-difference estimates and then performed logistic regression analysis using the same framework and set of regressors as in the cost models aforedescribed.

 Unlike in the cost regression analysis, the same sets of families are not eligible for each measure in the pre-post years. For the remainder of the HEDIS outcomes, we stratified the sample into those who did or did not receive the recommended care in 2004 and within strata compared the rates of receiving the recommended care in 2005 by treatment status, both unadjusted and controlling for the same set of regressors as theprior  models. The stratification was to address concerns that families who are about to transition into a high-deductible plan will try to obtain care that they anticipate needing before the transitions; that is, they will try to “stock up” on care.

Analyses were performed using commercially available statistical software (SAS version 9.1; SAS Institute, Cary, North Carolina; and STATA version 10; StataCorp Inc, College Station, Texas). We report statistical significance levels from 2-sided tests without adjustment for multiple testing. Full results are provided in the eAppendices.

Enrollees in high-deductible plans were more likely to be single men, were younger, had lower risk scores (better baseline health), and lived in areas with higher percentages of college graduates and non-Hispanic whites than families enrolled in control plans. After weighting using propensity scores, the samples have similar measured characteristics (

At baseline, both groups were enrolled in plans with actuarial values that averaged 82%. The baseline monthly costs of the treatment and weighted control groups are given in 

): both groups had similar monthly family healthcare costs of just over $500 and a distribution of costs by service type typical of those with employer-provided insurance. Before weighting, cost growth for the control group was 13%, similar to estimates for other data sources covering this period. Differences in the growth for the control group after weighting reflect the alignment of the controls to match those who enroll in treatment plans.

Effects of HDHP or CDHP Enrollment on Cost Growth

Costs grew for both the treatment families and the control families, but they grew more slowly in the higher-deductible group (Table 2). The monthly costs of the households enrolled in higher-deductible plans grew by $85 less than the comparable controls; in percentage terms, the expenditures of the control group grew by 20%, while the expenditures of the treatment group grew by 4%. Consequently, in the first year after enrolling in an HDHP or a CDHP, spending was 14% (95% confidence interval, 11.3%-16.9%) lower than that for comparable families in control plans (difference in the post-year mean monthly costs for treatment and control families divided by the mean post-year costs for control families). This was due to lower growth in inpatient, outpatient, and prescription drug costs. Growth in expenditures for emergency department care did not differ significantly between the 2 groups.

, cost growth for families in plans with moderately high deductibles ($500-$999) did not differ significantly from costs of those in control plans. However, cost reductions were greater (and significant) for families in plans with deductibles of $1000 or more. These cost reductions were maintained at a similar level when an account option was added with a low employer contribution (<$500 [mean, $399.32]). However, these cost reductions were attenuated when employers made generous contributions (>$500 [mean, $768.38]) to the accounts. This pattern of results across plan characteristics held for each of the individual care settings as well (discussed further in the eAppendices).

Some evidence was observed of increases in healthcare costs in the final quarter of 2004 among families who were about to enter an HDHP or a CDHP in 2005 (not  significant for the treatment group as a whole but significant for those with deductibles >$1000), suggesting possible stocking up by those about to change insurance. Because the opportunity for stocking up is limited by the timing of information on health plan offerings for the upcoming year, insurance restrictions on the frequency of many procedures, and uncertainty about future health needs, we assume that any stocking up of services would occur near the time of the insurance change and would include services that would otherwise be obtained early at the start of the new plan-year. As a robustness check (discussed further in the eAppendices), we compared cost growth using the same framework but using only costs from the second and third quarters of 2004 and 2005. We obtained similar results.

POLICY

Enrollment in high-deductible health plans is associated with reduced spending on healthcare and with moderate reductions in the use of preventive care.

Healthcare Spending and Preventive Care in High-Deductible and Consumer-Directed Health Plans

 To provide state-level estimates of the size and structure of the US individual market for health insurance and to investigate the potential impact of new medical loss ratio (MLR) regulation in 2011, as indicated by the Patient Protection and Affordable Care Act (PPACA).

 Using data from the National Association of Insurance Commissioners, we provided state-level estimates of the size and structure of the US individual market from 2002 to 2009. We estimated the number of insurers expected to have MLRs below the legislated minimum and their corresponding enrollment. In the case of noncompliant insurers exiting the market, we estimated the number of enrollees that may be vulnerable to major coverage disruption given poor health status.

 In 2009, using a PPACA-adjusted MLR definition, we estimated that 29% of insurer-state observations in the individual market would have MLRs below the 80% minimum, corresponding to 32% of total enrollment. Nine states would have at least one-half of their health insurers below the threshold. If insurers below the MLR threshold exit the market, major coverage disruption could occur for those in poor health; we estimated the range to be between 104,624 and 158,736 member-years.

 The introduction of MLR regulation as part of the PPACA has the potential to significantly affect the functioning of the individual market for health insurance.

	Estimation of the number of insurers expected to have medical loss ratios (MLRs) below the legislated minimum and their corresponding enrollment indicated the following:

		In 9 states, at least 50% of health insurers would likely fail to meet the 80% minimum MLR.

		In 12 states, at least one-half of total member-years of enrollment were affiliated with health insurers under the minimum threshold.

		If market exit is pursued by insurers not in compliance, coverage disruption could occur for those in poor health.

A central focus of the Patient Protection and Affordable Care Act (PPACA) of 2010 is to improve the functioning of the individual and small employer group markets for health insurance through increased regulation. One provision includes establishing minimum medical loss ratios (MLRs), which represent the percentage of a health insurer’s premium revenues that are paid out for clinical services. The PPACA establishes minimum MLRs of 80% for the small group (1-100 workers) and individual markets for health insurance, and 85% for the fully insured large group market beginning January 1, 2011.

 Insurers that have MLRs below the minimum threshold are required to provide a rebate to enrollees equal to an amount that reflects the premium revenue corresponding to the difference between its actual MLR and the minimum requirement.

A primary motivation behind this regulation is to ensure that premiums overwhelmingly reflect costs associated with enrollees’ receipt of clinical services, rather than excess  profitability or administrative costs that provide little direct value to consumers.

 However, multiple stakeholders are concerned about how insurers will respond to this new regulation and what, if any, consequences it may have for individuals’ access to coverage. Of concern is whether the individual market, which currently serves approximately 7% of individuals younger than 65 years, will become less stable (eg, insurer exit, closed blocks of business, increased barriers to access).

 Using annual filing data from the National Association of Insurance Commissioners (NAIC), we provide estimates of the size and structure of the individual market for health insurance, as well as MLRs across states over the 2002-2009 time period. With the 2009 data, we generated state-level estimates of the number of individual market insurers expected to fall below the minimum threshold and their corresponding enrollment. Finally, we estimated the amount of individual market enrollment associated with insurers falling below the minimum MLR threshold that may be vulnerable to major coverage disruption due to poor health status. We conclude with a discussion of how these findings can inform policymaking activities relating to implementation and individual market functioning. 

Our primary data source was the NAIC Statistical Compilation of Annual Statement Information for Health Insurance Companies for 2002-2009. (The NAIC annual statements are subject to both an annual audit from a certified public accountancy firm and an examination from the state at least every 5 years. Additionally, there is a dynamic of restating balance sheet items in various levels of detail and categorizations, which allows for cross-checking of data for reporting consistency. Finally, not only are the NAIC data reviewed by the domiciliary regulator, but also other states in which the insurer writes business may review a company’s reports.) Although these data are regularly used by state regulators and industry leaders, they are limited in 2 aspects. First, the vast majority of insurers operating within California are regulated by the California Department of Managed Health Care and do not file with the NAIC. Second, approximately 20% of premiums for comprehensive major medical policies in the individual market are written by life insurers, which do not file state-level information on enrollment, premiums, and claims specific to comprehensive major medical policies in the individual market. (For more details, please refer to 

This study investigates the potential impact of new medical loss ratio regulation on the individual market for health insurance in the United States.

March 2011

Fracture Risk Tool Validation in an Integrated Healthcare Delivery System

Quality of Anticoagulation Control Among Patients With Atrial Fibrillation

Healthcare Spending and Preventive Care in High-Deductible and Consumer-Directed Health Plans

Regulating the Medical Loss Ratio: Implications for the Individual Market

Improving Endoscopy Completion: Effectiveness of an Interactive Voice Response System

On Telephone-Based Disease Management: Wrong Diagnosis, Right Prescription

Connecting Statewide Health Information Technology Strategy to Payment Reform

Telemonitoring With Case Management for Seniors With Heart Failure

Health Information Technology and Physicians' Perceptions of Healthcare Quality

COPD-Related Healthcare Utilization and Costs After Discharge From a Hospitalization or Emergency Department Visit on a Regimen of Fluticasone Propionate-Salmeterol Combination Versus Other Maintenanc