n, PhD

To assess the cost-effectiveness of a community-based patient navigation program to improve cervical cancer screening among Hispanic women 18 or older in San Antonio, Texas. 

We used a microsimulation model of cervical cancer to project the long-term cost-effectiveness of a community-based patient navigation program compared with current practice. 

We used program data from 2012 to 2015 and published data from the existing literature as model input. Taking a societal perspective, we estimated the lifetime costs, life expectancy, and quality-adjusted life-years and conducted 2-way sensitivity analyses to account for parameter uncertainty. 

The patient navigation program resulted in a per-capita gain of 0.2 years of life expectancy. The program was highly cost-effective relative to no intervention (incremental cost-effectiveness ratio of $748). The program costs would have to increase up to 10 times from $311 for it not to be cost-effective. 

The 3-year community-based patient navigation program effectively increased cervical cancer screening uptake and adherence and improved the cost-effectiveness of the screening program for Hispanic women 18 years or older in San Antonio, Texas. Future research is needed to translate and disseminate the patient navigation program to other socioeconomic and demographic groups to test its robustness and design.

Community-based patient navigation programs may improve cervical cancer screening uptake, especially among Hispanic women. This study provides healthcare managers with knowledge about patient navigation programs that are multilevel and include some elements and principles from behavioral economics to improve cancer screening. 

This study adds to the existing literature by assessing the long-term cost-effectiveness of a community-based patient navigation program for Hispanic women. 

This study promotes the implementation of patient-centered cancer screening services.

Although advances in screening and vaccination technologies have substantially lowered the risk of cervical cancer among women, it still accounts for more than 4000 deaths per year in the United States.

 There are also persistent disparities in incidence and mortality rates, not only by socioeconomic status and geography, but also by ethnicity and race.

 Hispanic women have a higher risk of cervical cancer than other major ethnic/racial groups and are more likely to be diagnosed at a later stage.

 In urban areas, reducing cervical cancer morbidity and mortality is particularly difficult as the success of cancer prevention programs requires knowledge and self-control that few patients possess.

Patient navigation refers to “the support and guidance offered to persons with abnormal cancer screening or a new cancer diagnosis in accessing the cancer care system, overcoming barriers, and facilitating timely, quality care provided in a culturally sensitive manner.”

 In public health practice, patient navigation may include a variety of specific services and interventions, such as scheduling appointments with culturally sensitive caregivers, providing transportation or interpretation services, and assisting participants with childcare during scheduled appointments.

 The results of several randomized controlled trials have shown that patient navigation is effective in increasing patient satisfaction, decreasing the anxiety associated with screening processes and procedures, and improving cancer screening uptake and adherence.

 However, there is still limited evidence supporting the efficacy of patient navigation in improving patient outcomes over the long term or assessing the cost-effectiveness (CE) of specific patient navigation programs.

In this study, we explored the implementation and results from a community-based patient navigation program (designed to increase the cervical cancer screening [Pap test] rate) in San Antonio, Texas, for an underserved Hispanic female population 18 years or older. The program was multilevel and included some elements and principles from behavioral economics.

 Because the benefits of cervical cancer screening are hardly observed in the short term, we also used an evidence-based microsimulation model to assess improvements in long-term patient outcomes, and to evaluate the CE of the program versus the status quo. Finally, to ensure the robustness of the CE analysis, we conducted sensitivity analyses to assess several key cost and effectiveness parameters.

This study focused on a 3-year patient navigation program for cervical cancer screening implemented by the Bexar County Hospital District (University Health System) in San Antonio, Texas, from 2012 to 2015. The program targeted an urban female Hispanic population 18 years or older enrolled in CareLink, a financial assistance program for the uninsured population in San Antonio. This population has a particularly high risk for cervical cancer: in 2009, approximately 67% of women aged at least 18 years who were enrolled in CareLink had not had the recommended Pap test within the past 3 years.

 In addition, these women faced a range of cultural and socioeconomic barriers to undergoing cancer screening (eg, lack of financial resources to access screening services, fear of embarrassment, and concerns about provider sensitivity to patient comfort).

The patient navigation program is recognized as a major component of a community-based, culturally competent, secondary cancer prevention program described in a previous study.

 It was designed to provide personalized social communication by encouraging participants to call “Claudia,” a bilingual female contact person who would act as a program navigator in disseminating health information. This is consistent with the behavioral economics principle of relying on social and cultural norms because using the same Hispanic name as a contact person helps participants recall similar events in memory in a culturally competent way.

 The program also included participant reminders to call Claudia within newsletters, public service announcements, and automated messages. Afterward, these patient navigators provided assessments of the cervical cancer and screening knowledge of patients they had spoken with, as well as personalized education about the potential benefits of screening.

In addition to the services provided by patient navigators, the program also implemented multilevel strategies designed to increase the uptake and adherence of cancer screening within the target population. For example, the program relied on a mass media health promotion campaign, which allowed women to align their subjective assessment of cervical cancer risk with their actual risk by receiving health education and information messages provided by patient navigators who are similar to, or representative of, the target population. The program also provided patients with accurate information related to cervical cancer risk to address unrealistic expectations (ie, individuals may have unreasonably low or high estimates of their cervical cancer risk).

 Lastly, as an incentive to each program participant, all screening tests were free. The patient navigation program was designed to be multilevel and to integrate general principles of behavioral economics by taking into account key factors that patients consider when making screening decisions.

The patient navigation program demonstrated its effectiveness at improving cervical cancer screening through interviews and focus groups that took place between program staff members and participants. In particular, 94% of program staff (including patient navigators, care providers, and others who participated in the implementation of the program) agreed that it had addressed the needs of cervical cancer screening among Hispanic women and participants were either very satisfied or satisfied working with the program. In addition, patients reported a positive experience using the program services provided, including increased knowledge about cervical cancer and HPV and stronger motivation to participate in cancer screenings. Overall, the program has navigated 4500 women in the target population and increased the 3-year screening rate from 65% to 80% during the 3-year study period. 

Although empirical studies based on actual data may produce important insights into the CE of a given patient navigation program, they are costly or may not be able to assess the long-term impact of cervical cancer screening strategies.

 Simulation modeling—particularly microsimulation models&mdash;offers a more flexible, cost-effective approach to conducting economic evaluations and making informed decisions compared with studies based on actual behavioral observations.

 By incorporating the best available biological, clinical, and epidemiological evidence, a microsimulation model of cervical cancer enables researchers to simulate a population of interest, capture the disease progression of each individual, predict the long-term consequences of different interventions within a virtual environment, and provide insights into the CE of different strategies designed for cervical cancer prevention.

Our model structure is as follows (a detailed model description can be found in the 

]): our microsimulation model incorporates state-of-the-art knowledge from previous cervical cancer decision models, and estimated parameters based on our specific program.

 The natural history of cervical cancer was modeled using 16 states, including: well (healthy) HPV infection, low- and high-grade squamous intraepithelial lesions (SILs), hysterectomy for benign disease, undetected and detected cervical cancer stages I to IV, survival from cancer, and death due to cervical cancer or other causes (

). Transitions between health states were governed by transition probabilities that were dependent on age, SILs level, cancer stage, and screening or vaccination strategies. The basic cycle length was 1 year.

Each year, women in the simulation model could be infected with HPV or stay uninfected. We assumed all cases of cervical cancer start with HPV infection, which is consistent with the epidemiologic finding that HPV causes a majority of cervical cancer cases.

 HPV infection, clearance, and progression to low- or high-grade SILs is a complex process that varies depending on HPV virus type and patient characteristics, such as age and immune status. We used average transition probabilities for all virus types (ie, we did not distinguish between different types of HPV), which simplified our model without losing important information. The incidence of HPV infection was modeled to be a function of age, and the parameters of the incidence function did not change throughout the simulation.

Women infected with HPV could regress to well, stay unchanged, or progress to low- or high-grade SILs. Similarly, women with low- or high-grade SILs could undergo regression, no change, or progression to stage I cancer without symptoms. Current knowledge about the natural history of cervical cancer suggests that most HPV infections will regress on their own and some persistent HPV infections may progress to high-grade SILs and cancer.

Women with asymptomatic stage I cancer either become symptomatic or progress to higher stages without detection. Once cancer becomes symptomatic or is detected by screening, the patient will undergo medical treatment. Women without cancer may undergo a hysterectomy due to other causes

 and all women could die from causes outside those identified in the study.

 summarize all input parameters in the microsimulation model. We estimated the incidence of HPV and transition probabilities among different health states from the published literature

 and age-specific female mortality rates from other causes by subtracting the rates due to cervical cancer from age-specific, all-cause mortality rates obtained from 2010 US life tables.

 Quality of life weights were determined by either age (for women without cervical cancer) or cancer stage (women with).

We used a societal perspective in the CEA by incorporating both program and treatment costs; the patient navigation and screening program costs were estimated to be $311 per person. We calculated this figure by adding all costs incurred in the program ($1,399,815), including navigation and screening-related program staff salaries, health promotion media and outreach costs, and Pap test cost, then dividing the total by the number of women (4500) who received patient navigation and screening services. We estimated annual treatment costs for different cancer stages from the published literature.

 Women with HPV infection or low-grade SILs do not need treatment, and thus, did not incur additional costs.

Lastly, the sensitivity and specificity of the screening test were estimated to be 80% and 95%, respectively, based on the published literature.

 Although we expected that the prevalence of HPV infection in the San Antonio metropolitan area would be higher than the national average, we still used the national average (26.8%) in our model due to a lack of population-specific data in our community of interest.

 All input data and parameters can be found in the eAppendix.

Our model followed 100,000 simulated women with the same age distribution and prevalence of HPV infection as the population of interest (Hispanic women 18 years or older) throughout their lifetime, with and without the patient navigation program. Simulating a large population provides stable estimates of long-term outcomes for each simulation scenario. In the CEA, we assumed that women who were successfully navigated through the study received Pap tests at suitable intervals, appropriate diagnostic procedures (eg, colposcopy, biopsy), and treatment based on the results of screening. Specifically, women with low-grade SILs were reexamined every 6 to 12 months until they had 3 negative screening test results.

 In addition, women with confirmed high-grade SILs or cancer were treated according to published guidelines.

We tracked the overall costs, life expectancy, and quality-adjusted life-years (QALYs) of the simulated population for each scenario and discounted them by 3% (a widely accepted discount rate in CEA) annually. We then measured the performance of the program by estimating the incremental cost-effectiveness ratio (ICER) between the patient navigation program and no intervention. We implemented the microsimulation model using the software package AnyLogic 7.1 (AnyLogic North America; Chicago, Illinois).

presents the CE estimates of the patient navigation program for cervical cancer screening compared with no intervention. Our results show that the program costs an average of $45 more per person than the no intervention scenario. Also, the screening program showed an increase in the life expectancy of the studied population by 0.2 years and an increase in QALYs by 0.06 years, which results in an ICER of $748 per QALY for the patient navigation program versus no intervention. We used $50,000 per QALY as the CE threshold to determine that the patient navigation program was considered to be cost-effective in the baseline scenario.

We studied the robustness of the baseline results by conducting sensitivity analyses that accounted for uncertainties in the cost and effectiveness of the patient navigation program. Figure 2 reports the results of a 2-way sensitivity analysis of the program cost per participant and screening rate. Again, we used $50,000 per QALY as the CE threshold. Specifically, if a combination of program cost per participant and the screening rate falls below the CE frontier shown in 

, the ICER of the patient navigation program relative to the status quo is less than $50,000 per QALY, which means the program is more cost-effective; otherwise, the status quo is more cost-effective. The results show that, because 80% of program participants received cervical cancer screening tests, the cost of the program could increase up to 10 times—from $311 to $3312&mdash;before the program becomes less cost-effective. In addition, even if the patient navigation program only resulted in a screening rate of 70% (a 5% increase from the status quo), the program is considered cost-effective as long as its price tag remains below $2353 per person.

Large disparities in cancer screening uptake and outcomes exist across many socioeconomic and demographic groups in the United States and, despite substantial progress to reduce these differences by developing new cancer screening initiatives, these gaps in cancer screening stubbornly persist.

 Community-based, multilevel patient navigation programs have shown promise in improving adherence to cancer screening processes and protocols. When these programs can further incorporate some principles of behavioral economics—with a focus on understanding the heuristics individuals use to make decisions&mdash;they can address patient biases in cancer risk and decision making surrounding cancer screening and optimize the appropriate architecture for individuals who are considering undergoing cancer screening.

Our results showed that a specific community-based patient navigation program for cervical cancer screening was cost-effective in increasing the screening rate and improving the long-term health outcomes of the target population. We estimated that an average program participant would gain an additional life expectancy of 0.2 years and an additional 0.06 QALYs. Under the baseline scenario, the patient navigation program costs $748 for each additional QALY gained with respect to no intervention, indicating that the program is highly cost-effective. Sensitivity analyses showed the robustness of the CE of the patient navigation program: 1) the program costs would have to increase up to 10 times from $311 before the program ceased to be cost-effective, and 2) the program would be cost-effective even if the screening rate only increased from 65% to 70% instead of the observed 80% after program implementation.

Similar community-based patient navigation programs have been shown to improve the cervical cancer screening rate of Hispanic women and the colorectal cancer (CRC) screening rate of Hispanic men. (In a previous study, we showed that a patient navigation program increased the CRC screening rate among target Hispanic men from 16% to 80% in data collected from 2011-2013.)

 The navigation program would reduce the lifetime overall cost to its participants (due to significantly reduced cancer risk) and thus achieve cost-savings compared with no intervention.

 Given these promising results, healthcare providers may consider testing and evaluating similar patient navigation programs to improve screening for other types of cancer (eg, breast cancer and prostate cancer) in other populations of interest.

The cervical cancer natural history model was developed based on parameters that reflect the general US population, not specifically the Hispanic population, which was the target population of this study. We also did not have local cost estimates for cervical cancer treatment; thus, we relied on national average costs estimated from the existing literature. We will update the CE results as more local data for model parameterization become available and conduct more comprehensive sensitivity analyses to address these parameter uncertainties.

We did not model the way in which different population characteristics or socioeconomic factors would influence the choice of screening in the model. One of our next steps will be to model a decision-making process for each individual participating in the simulation regarding whether to undergo screening, which would increase the realism of the model by considering individual heterogeneity and potentially improve the validity of the CE results.

Finally, we did not examine the effect of varying screening intervals or HPV vaccination on the projected outcomes. Modeling these additional scenarios is a next step, as our model can be easily adapted to incorporate a different screening interval or include HPV vaccination as part of cervical cancer prevention practices.

Our study results demonstrate how a health system serving a low-income, urban, and minority (Hispanic) population was able to develop a cost-effective patient navigation program for cervical cancer screening. Although our findings are promising, the patient navigation program results presented here need to be translated and disseminated to other socioeconomic and demographic groups to test the robustness and design of the program, particularly in terms of how to carefully calibrate behavioral change components and understand which program attributes are most promising.

The authors thank David Siscovick, MD, MPH, for his constructive comments.

 Center for Health Innovation, The New York Academy of Medicine (YL, JAP), New York, NY; Department of Population Health Science and Policy, Icahn School of Medicine at Mount Sinai (YL, JAP), New York, NY; College of Nursing and Health Innovation, The University of Texas at Arlington (EC), Arlington, TX; Research and Information Management, University Health System (RV, LM), San Antonio, TX; Leonard Davis Institute of Health Economics, University of Pennsylvania (JAP), Philadelphia, PA.

 This study was funded by the Cancer Prevention and Research Institute of Texas (award grant ID PP120111).

 The authors 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 (YL, JAP); acquisition of data (YL, RV, EC, LM); analysis and interpretation of data (YL, JAP); drafting of the manuscript (YL); critical revision of the manuscript for important intellectual content (YL, RV, LM, JAP); statistical analysis provision of patients or study materials (EC); obtaining funding (RV); and administrative, technical, or logistic support (EC, LM).

 Yan Li, PhD, Center for Health Innovation, The New York Academy of Medicine, 1216 Fifth Ave, New York, NY 10029. E-mail: yli@nyam.org. 

1. Siegel R, Naishadham D, Jemal A. Cancer statistics, 2013. 

. 2013;63(1):11-30. doi: 10.3322/caac.21166.

2. Saraiya M, Ahmed F, Krishnan S, Richards TB, Unger ER, Lawson HW. Cervical cancer incidence in a prevaccine era in the United States, 1998—2002. 

3. Singh GK, Miller BA, Hankey BF, Edwards BK. Persistent area socioeconomic disparities in US incidence of cervical cancer, mortality, stage, and survival, 1975—2000. 

4. Siegel R, Naishadham D, Jemal A. Cancer statistics for Hispanics/Latinos, 2012. 

. 2012;62(5):283-298. doi: 10.3322/caac.21153.

5. Bastani R, Berman BA, Belin TR, et al. Increasing cervical cancer screening among underserved women in a large urban county health system: can it be done? what does it take? 

6. Freund KM, Battaglia TA, Calhoun E, et al; Patient Navigation Research Program Group. National Cancer Institute Patient Navigation Research Program: methods, protocol, and measures. 

. 2008;113(12):3391-3399. doi: 10.1002/cncr.23960.

7. Percac-Lima S, Grant RW, Green AR, et al. A culturally tailored navigator program for colorectal cancer screening in a community health center: a randomized, controlled trial. 

. 2009;24(2):211-217. doi: 10.1007/s11606-008-0864-x.

Articles

This study investigates the cost-effectiveness of a community-based patient navigation program to improve cervical cancer screening.

Cost-Effectiveness of a Patient Navigation Program to Improve Cervical Cancer Screening

To determine the impact of individual features of the patientcentered medical home (PCMH) care model on next-year healthcare expenditures including outpatient, inpatient, emergency department, pharmacy, and total healthcare expenditures among Medicare beneficiaries 65 years and older.

Analysis of retrospective longitudinal survey data. Methods Longitudinal files from the Medical Expenditure Panel Survey were analyzed. Differences in expenditures for individuals whose usual sources of care did or did not have different PCMH features were estimated using recycled predictions from generalized linear regression models.

Having little to no difficulty contacting the regular source of care over the telephone during regular business hours was associated with significantly lower total and inpatient expenditures over the next year (differences of $2867 and $3736, respectively). Having a regular source of care with office hours at night or on weekends was associated with significantly lower outpatient, emergency department, and other expenditures (differences of $535, $103, and $328, respectively). Pharmacy expenditures were significantly higher for individuals whose usual source of care inquired about medications and treatments prescribed by other doctors (difference of $362).

This study points out the need to identify how individual PCMH features impact healthcare expenditures across different policy-relevant categories. Practices that have not fully adopted a PCMH model can still make progress in improving quality and controlling costs by adopting even some modest features of the PCMH model.

Features of the patient-centered medical home (PCMH) care model individually influence future healthcare expenditures among Medicare beneficiaries.

Total, inpatient, outpatient, emergency department, pharmacy, and other expenditures were differentially affected by 1 or more individual features of PCMH care.

This study points out the need to identify how individual PCMH features impact healthcare expenditures across different policy-relevant categories.

Results revealed that practices that have not fully adopted a PCMH model can still make progress in improving healthcare quality while reducing or controlling costs by adopting even some modest features of the PCMH model.

Patient-centered medical homes (PCMHs) are showing promise as a novel way to improve healthcare quality while keeping healthcare cost growth under control.

 Through coordinated, team-based approaches to healthcare delivery that are tailored to address the needs of individual patients via enhanced communication, PCMHs shift the focus of healthcare delivery from the system level to the patient level.

 PCMH models have been implemented in single healthcare systems, and studies of these interventions and demonstrations have focused on implementation costs, patient experiences, evidence-based care processes, specific health outcomes, and healthcare utilization and costs. Published studies indicate that PCMHs are associated with small improvements in overall patient satisfaction with care and reported satisfaction with care coordination and communication,

as well as moderate enhancements to clinical care delivery and processes, primarily for preventive services.

 There is also some evidence of potential associations of PCMHs with improved glycated hemoglobin and low-density lipoprotein values,

as well as decreased short-term mortality rates among older adults.

Studies to date indicate small improvements in inpatient and emergency department utilization among patients engaged with PCMHs,

but none show significant cost savings associated with PCMHs.

 However, from health policy or managed care perspectives (eg, a third-party payer or an accountable care organization), it is unclear how PCMHs impact healthcare expenditures across different levels of care (eg, outpatient care, emergency department [ED] or inpatient care).

All PCMHs deliver care by combining a set of different features, components, or services that complement each other, with the goals of enhancing care delivery and communication. For example, the National Committee for Quality Assurance’s (NCQA's) Physician Practice Connections—Patient- Centered Medical Homes recognition program includes 9 standards addressing areas such as access and communication, referral tracking, and performance reporting and improvement.

 Another example is the Comprehensive Primary Care (CPC) initiative by the Centers for Medicare & Medicaid Services (CMS).

 Under the CPC initiative, the PCMH model is augmented by multipayer payment reform (eg, by offering bonus payments to doctors who improve care coordination), total cost accountability in the form of shared savings, and the requirement that the 500 participating primary care practices (serving 313,000 Medicare beneficiaries) use electronic health records (EHRs) to better coordinate care.

Little is known about the role of different individual components that define a PCMH on explaining variation in future healthcare expenditures. This is important because the costs of implementing a PCMH model in a medical practice are nontrivial and, as a result, medical practices have to decide which elements should be implemented first. In the context of managed care, it is also important to know which PCMH features disrupt all the different categories of healthcare expenditures—which will impact profits, particularly under shared savings arrangements. Although CMS is in the process of testing Medicare PCMH models,

 there is currently a lack of PCMH research on the Medicare population. In this study, we use data from the Medical Expenditure Panel Survey (MEPS) to determine the impact of individual features of the PCMH model on different levels of future healthcare expenditures, including outpatient, inpatient, ED, pharmacy, and total healthcare costs among Medicare beneficiaries 65 years and older.

The Household Component of the MEPS was the data source for this study. The Agency for Healthcare Research and Quality administers the MEPS, collecting in-depth information about annual healthcare utilization, medical expenditures and health conditions from a sample of households in the United States. MEPS employs an overlapping panel design; a new panel of sample households is selected each year and then tracked over the 2-year period. The 3 most recent MEPS Longitudinal Data Files were utilized, which included Panels 12, 13, and 14, interviewed over the years 2007 to 2010.

The study sample included adults who were 65 years of age or older, indicated that they were enrolled in Medicare, and reported that they had a usual source of care other than the ED. Analyses were limited to adults who were not missing data on variables of interest. As a result, 2387 individuals qualified for all analyses (54.8% of the study sample). The most common questions with missing data were those addressing PCMH features (eg, 1920 of the 1970 respondents with missing data did not include information on 1 or more of the questions used to identify PCMH features). The sample size was sufficient for determining statistical significance.

Expenditure variables describing the total of payments for care during the second year of each 2-year panel in total, and by health service category, were the outcome variables of interest. These expenditure variables were based on the sum of expenditures during the year from all payment sources, including out-of-pocket payments and payments by third-party payers. The health services expenditure categories included in the current analysis were outpatient (including office-based and hospital outpatient visits), inpatient, ED, prescription medication, and other (including dental care, home healthcare, vision aids, and other medical supplies and equipment).

Five variables from second-round interviews describing the features of a PCMH as described by Beal and colleagues15 were the primary independent variables for this study. These questions were worded as follows: (1) “How difficult is it to contact (a medical person at) (PROVIDER) during regular business hours over the telephone about a health problem?”; (2) “Does (PROVIDER) have office hours at night or on weekends?”; (3) “How difficult is it to contact (a medical person at) (PROVIDER) after their regular hours in case of urgent medical needs?”; (4) “Does (someone at) (PROVIDER) usually ask about prescription medications and treatments other doctors may give them?”; and (5) “If there were a choice between treatments, how often would (a medical person at) (PROVIDER) ask (you/name) to help make the decision?”

Responses of “yes,” “no,” “refused,” and “don’t know” were analyzed such that “yes” was coded as a positive response (1), “no” was coded as negative (0), and “refused” and “don’t know” were coded as missing. Questions asking about difficulties were dichotomized, with “not too difficult” and “not at all difficult” coded as positive (1), “refused” and “don’t know” coded as missing, and the remaining responses coded as negative response (0). The frequency question was also dichotomized, with “usually” and “always” coded as positive (1), “refused” and “don’t know” coded as missing, and the remaining responses coded as negative (0).

The Beal study, which defined the elements of a PCMH for the current study, also included 3 other variables that help define PCMHs.

 These 3 variables looked at whether individuals visited their regular source of care for new problems, preventive care, and ongoing health problems. They were not included in the current study because most of the responses for these 3 variables were positive (98.8%, 98.7%, and 98.0%, respectively).

Additional variables were incorporated to describe the population and adjust for potential confounders of the relationship between PCMH features and healthcare expenditures. With the exception of race/ethnicity (which is assessed during the initial round) and household income (which is based on income for the first year of the panel), these additional variables were based on the responses to questions asked during second-round interviews. These variables included age, race/ethnicity, region, marital status, poverty status, health insurance coverage, activities of daily living (ADLs) and instrumental activities of daily living (IADLs) limitations, chronic health conditions, and perceived health status). Categorizations for these variables are described in 

We first explored the characteristics of the population and investigated how the potential confounding variables differed across the different features of a PCMH. χ² tests were used to assess the significance of unadjusted differences between the categories in each of the PCMH features. Next, we conducted analyses of second-year expenditures associated with each of the PCMH features. As the second-year expenditures for Panels 12, 13, and 14 represented expenditures in 2008, 2009, and 2010, respectively, the expenditures for Panels 12 and 13 were adjusted for inflation to 2010 levels of expenditures using the Consumer Price Index for Medical Care Services.

Then, for each of the expenditure categories, we calculated the unadjusted and adjusted predicted expenditures associated with each of the PCMH features using generalized linear regression models with a gamma distribution and a loglink function. This statistical approach accounted for the skewed nature of cost data.

 Unadjusted models included only a single PCMH feature as a predictor, while adjusted models included all PCMH features and all potential confounders described previously.

The average differences in costs for individuals who had usual sources of care with and without each of the PCMH features were determined by predicting healthcare expenditures using the estimated coefficients from the generalized linear regression equations. This yielded predictions for each individual based on the typical expenditures for individuals with similar characteristics, which were estimated using the method of recycled predictions.16 We then calculated the differences in these average expenditures. All statistical analyses took into account the complex survey design of MEPS and were conducted in Stata MP version 12.1 (StataCorp LP, College Station, Texas).

 describes the percentage of the study population (N = 2387) by the reported PCMH features. The most common PCMH feature was always or usually asking patients to help decide between treatments, with 85.4% of respondents reporting this feature (95% confidence interval [CI], 83.1-87.6). The least common PCMH feature was having office hours on nights or weekends, with 29.6% of respondents reporting this feature (95% CI, 26.4-32.7). The reported frequency of the other features are provided in Table 1, and information about how the PCMH features vary by sociodemographic and health status variables is provided in eAppendix A.

 reports the unadjusted and adjusted average expenditures associated with each of the PCMH features (see

for CIs). Having little to no difficulty contacting the regular source of care over the telephone during regular business hours was associated with significantly lower total expenditures and inpatient expenditures. The adjusted average inpatient expenditures for individuals who had little to no difficulty contacting their regular source of care by telephone during business hours was $3230, while the adjusted average inpatient expenditures for individuals who reported difficulty was $6966, a difference of $3736 (

 = .0177). Total adjusted average healthcare expenditures were also significantly lower for individuals who had no difficulty contacting their regular source of care by telephone during business hours. The adjusted average total healthcare expenditures for individuals who had little difficulty contacting their regular source of care during business hours was $10,117, while the adjusted average total expenditures for individuals who had difficulty was $12,984, a difference of $2867 (

On the other hand, there were no significant differences in the adjusted average annual medical expenditures for outpatient services (difference of —$888, 

 = .0852), ED services (difference of —$59, 

 = .4443), pharmacy services (difference of $94,

 = .6782), or other health services (difference of $195, 

 = .3122) depending on whether or not the patient had difficulty contacting his or her usual source of care by telephone during business hours.

Having a regular source of care with office hours at night or on the weekend was associated with significantly lower outpatient, ED, and other healthcare expenditures. The adjusted average outpatient expenditures for individuals who had a usual source of care with office hours at night or on the weekend was $2531, while the adjusted average outpatient expenditures for individuals without a usual source of care with this feature was $3066, a difference of $535 (

 = .0201). The adjusted average ED expenditures for individuals who had a usual source of care with office hours at night or on the weekend was $195, while the adjusted average ED expenditures for individuals without a usual source of care with this feature was $298, a difference of $103 (

 = .0234). The adjusted average for other health expenditures for individuals who had a usual source of care with office hours at night or on the weekend was $1057, while the adjusted average for other health expenditures for individuals without a usual source of care with this feature was $1386, a difference of $328 (

 = .0181). However, there were no significant differences in the adjusted average annual medical expenditures in total (difference of —$1485, 

 = .065) or for inpatient services (difference of —$879, 

 = .2028) or pharmacy services (difference of $34, 

 = .8693) depending on whether or not the usual source of care had office hours at night or on the weekend.

Having no difficulty contacting the usual source-ofcare provider after regular hours was not associated with significant differences in total or in any individual health service category. There were no significant differences in the adjusted average annual medical expenditures for total health services (difference of $967, 

 = .2334), outpatient services (difference of —$84, 

= .7905), inpatient services (difference of $1380, 

= .0665), ED services (difference of —$15, 

= .8012), pharmacy services (difference of $169, 

= .3908), or other health services (difference of —$122,

 = .5349) depending on whether or not the patient had difficulty contacting his or her usual source-of-care provider after regular hours.

There were no statistically significant differences in the adjusted average annual medical expenditures for total health services (difference of —$279, 

 = .7729), outpatient services (difference of $23, 

 = .9311), inpatient services (difference of —$360, 

= .7068), ED services (difference of —$41, 

 = .5252), or other health services (difference of $141, 

 = .423) depending on whether or not the usual source of care asked about medications and treatments prescribed by other doctors. However, the average pharmacy expenditures differed significantly depending on whether or not the usual source of care asked about medications and treatments prescribed by other doctors. The adjusted average pharmacy expenditures for individuals whose usual source of care asked about medications and treatments prescribed by other doctors was $2735, while the adjusted average pharmacy expenditures for individuals whose usual source of care did not ask about medications and treatment prescribed by other doctors was $2372, a difference of $362 (

Having a usual source of care that always or usually asks the patient to help decide between treatments was not associated with significant differences in total expenditures or in any individual health service category. There were no significant differences in the adjusted average annual medical expenditures for total health services (difference of $3, 

 = .9974), outpatient services (difference of —$8, 

 = .978), inpatient services (difference of $151, 

 = .8436), ED services (difference of $64, 

= .3239), pharmacy services (difference of —$337, 

 = .2284), or other health services (difference of —$116, 

 = .6173) depending on whether or not the patient was asked to help decide between treatments.

Having little to no difficulty contacting a regular source of care over the telephone during regular business hours was associated with significantly lower total healthcare expenditures as well as lower inpatient expenditures, which may be related to reduced hospitalizations. However, there were no significant differences in adjusted average annual healthcare expenditures for outpatient, ED, pharmacy, and other health services. Having a regular source of care with office hours at night or on the weekend was associated with significantly lower outpatient, ED, and other healthcare expenditures—perhaps due to a reduction in ED use&mdash;but it was not associated with significant differences in the adjusted average annual healthcare expenditures for inpatient or pharmacy services.

Having no difficulty contacting the usual source-ofcare provider after regular hours was not associated with significant differences in total expenditures or in any individual health service category. There were also no statistically significant differences in the adjusted average annual healthcare expenditures for outpatient, inpatient, ED, and other health services depending on whether or not the usual source of care asked about medications and treatments prescribed by other doctors.

Average pharmacy expenditures differed significantly depending on whether or not the usual source of care asked about medications and treatments prescribed by other doctors. This finding may be related to better medication management. For example, when a doctor asks about medication and treatments prescribed by other doctors, then this may be a marker for better medication management; the end result is that prescription expenditures will be higher, but all other expenditures may remain the same. Lastly, having a usual source of care that always or usually asks the patient to help decide between treatments was not associated with significant differences in total expenditures or in any individual health service category.

This study has several limitations. Some features of PCMHs could not be included in the current study because MEPS did not ask questions related to all the relevant PCMH features (eg, adoption of EHRs, use of evidence-based clinical protocols, degree of care coordination when transitioning between different levels of care). However, this study clearly points out the need to identify how individual PCMH features or components impact healthcare expenditures across different policyrelevant categories to better manage care. More specifically, the PCMH Recognition Program by the NCQA allows medical practices to qualify for financial incentives from health insurance plans and employers. Recognition is determined by achieving a set of standards related to, for example, patient access, communication, data tracking, and performance reporting.

 CMS also has developed programs to reward practices that fully implement PCMH models (eg, the CPC Initiative).

 Although studies have shown that PCMHs can improve care coordination and communication,

this study reveals that practices that have not fully adopted a PCMH model can still make progress in improving healthcare quality while reducing or controlling costs if they adopted even some modest features of a PCMH model. For example, Medicare could consider incentivizing medical practices that adopt individual cost-saving features of PCMHs and, thus, facilitate the progression of these medical practices toward a fully implemented PCMH model. Although the precise identification of cost-saving PCMH features remains a challenge, given data constraints, this may be the only financially feasible path available for many medical practices that are unable to fully absorb the costs associated with PCMH implementation. Future research is needed to more precisely identify and track how PCMH features related to the use of EHRs and performance reporting are directly related to quality and cost over the long term.

 University of North Texas Health Science Center, School of Public Health, Department of Health Management and Policy, Fort Worth, TX (ELS, LMP); Magellan Health Services, Analytic Services Department, Maryland Heights, MO (ELS); Baylor Scott & White Health, Office of the Chief Quality Officer, Dallas, TX (LMP); Center for Health Innovation, The New York Academy of Medicine, New York, NY (JAP); Leonard Davis Institute of Health Economics, University of Pennsylvania, Philadelphia, PA (JAP).

Concept and design (ELS, LMP, JAP); analysis and interpretation of data (ELS, LMP, JAP); drafting of the manuscript (ELS, LMP, JAP); critical revision of the manuscript for important intellectual content (ELS, LMP, JAP); statistical analysis (ELS, LMP); administrative, technical, or logistic support (ELS).

 Erica L. Stockbridge, Department of Health Management and Policy, School of Public Health, University of North Texas Health Science Center, 3500 Camp Bowie Blvd, Fort Worth, TX 76107. E-mail: els0127@live.unthsc.edu.1. Jackson GL, Powers BJ, Chatterjee R, et al. The patient-centered medical home: a systematic review. Ann Intern Med. 2013;158(3): 169-178.

2. Solberg LI, Asche SE, Fontaine P, Flottemesch TJ, Anderson LH. Trends in quality during medical home transformation. Ann Fam Med. 2011;9(6):515-521.

3. Jaen CR, Ferrer RL, Miller WL, et al. Patient outcomes at 26 months in the patient-centered medical home National Demonstration Project. Ann Fam Med. 2010;(8, suppl 1):S57-S67, S92.

4. Reid RJ, Fishman PA, Yu O, et al. Patient-centered medical home demonstration: a prospective, quasi-experimental, before and after evaluation. Am J Manag Care. 2009;15(9):e71-e87.

5. Wise CG, Bahl V, Mitchell R, West BT, Carli T. Population-based medical and disease management: an evaluation of cost and quality. Dis Manag. 2006;9(1):45-55.

Analysis of the impact of individual features of the patient-centered medical home care model on future healthcare expenditures among Medicare beneficiaries.

n, PhD

Articles

Cost-Effectiveness of a Patient Navigation Program to Improve Cervical Cancer Screening

Patient-Centered Medical Home Features and Expenditures by Medicare Beneficiaries