David T. Liss, PhD

Appropriate vaccinations are important for patients with rheumatoid arthritis (RA), who are often treated with highly immunosuppressive therapies that increase their risk of infection. However, rates of vaccination among patients with RA are below optimal levels.

We conducted a patient survey to assess self-reported vaccination status and to compare that status with electronic health record (EHR) data.

We recruited randomly selected patients with RA in an academic practice in 2013. Eligible participants had a diagnosis of RA, at least 1 visit to a rheumatology clinic in each of the previous 2 years, were 18 years or older, and had English listed as their preferred language. The survey included the following domains: a) patient self-reported receipt of influenza, pneumococcal (PNVX), and herpes zoster (HZVX) vaccinations; b) attitudes about these vaccines, including reasons for unvaccinated status, if applicable; and c) provider recommendations about these vaccines.

Based on participants’ self-report, we found a high vaccination rate for influenza during the previous season (79.4%), a moderate rate of any previous vaccination for pneumococcus (53.9%), and a very low rate of any previous vaccination for herpes zoster (7.8%). If we assume that all self-reports are accurate and we include vaccinations recorded in the EHR that were not reported by patients, the vaccination rates were approximately 8% to 9% higher for PNVX and HZVX.

Vaccination rates are low among patients with RA based on self-report data. Further research is needed to investigate system-level barriers to vaccination and the impact of evidence-based, provider-level interventions on vaccination rates. 

	Vaccination rates are low among patients with rheumatoid arthritis based on self-report data. 

		Lack of physician recommendation, as recalled by the patient, appears to be the most important factor for unvaccinated status. Only three-fourths of patients said their doctors had discussed their increased risk of infection, and only two-thirds said their doctors discussed the importance of vaccinations. 

		Patient lack of knowledge or negative attitudes toward vaccinations appear to be less important contributors to suboptimal vaccination rates. 

		Further research is needed to investigate system-level barriers to vaccination and the impact of evidence-based provider-level interventions on vaccination rates.

Patients with autoimmune rheumatic diseases have a risk of infection approximately double that of age- and gender-matched controls.

 Patients with rheumatoid arthritis (RA) are at an increased risk for infection due both to inherent dysregulation of the immune system and, for many, the added effect of chronic immunosuppressive therapy. For example, the adjusted hazard ratio for contracting herpes zoster in patients with RA is 1.9 compared with healthy controls.

Because of their increased risk of infection, the Advisory Committee on Immunization Practices (ACIP) recommends that immunocompromised adults receive an annual influenza vaccination (INFVX), as well as the pneumococcal vaccination (PNVX) with the 13-strain pneumococcal conjugated vaccine (PCV-13) followed by the 23-strain pneumococcal polysaccharide vaccine (PPSV-23).

 The recommendations are more complex for herpes zoster vaccination (HZVX), which uses a live virus and has historically been considered contraindicated in patients who are severely immunosuppressed. Currently, ACIP recommends HZVX for all individuals 60 years or older.

 However, a recent study found that HZVX had even higher efficacy (ie, 70% risk reduction) among healthy adults aged 50 to 59 years

; HZVX (Zostavax) has FDA approval for use in adults 50 years or older.

 ACIP states that HZVX is safe for patients who are taking less than 20 mg/day of prednisone and/or low doses of methotrexate (≤0.4 mg/kg/week), azathioprine (≤3.0 mg/kg/day) or 6-mercaptopurine (≤1.5 mg/kg/day).

 If a physician plans to initiate a strong immunosuppressant, ACIP recommends the patient be given the HZVX at least 14 days prior to starting the medication.

 This recommendation has been challenged by findings from a recent analysis of Medicare patients with immune-mediated disease who were given HZVX. There was no increase in herpes zoster infections during the 42 days after HZVX, even among patients receiving biologic immunosuppression, whereas those vaccinated had a lower incidence of zoster over a median 2-year follow-up.

Vaccination rates remain low among RA patients, despite their high risk of infection, strong evidence of vaccine safety and efficacy, and recommendations from national organizations. A recent large study found that only 25% of RA patients were adherent to current vaccine recommendations.

 Studies have shown adherence rates among RA patients ranging from 10% to 34% for INFVX,

 The reasons for these low vaccination rates are not entirely clear.

At our institution, the Northwestern Memorial Medical Center in Chicago, vaccination rates for patients with RA measured using electronic health record (EHR) data have been similar to national data. However, EHR vaccination data may be incomplete because many patients receive vaccinations outside of their rheumatologist’s office and their primary care clinician may practice in a different health system. The completeness of EHR data on vaccinations is also limited, as many patients may get vaccinated in pharmacies and retail stores with no automated way to transfer documentation to their healthcare team.

 Therefore, at the start of a quality improvement research project to improve vaccination rates, we conducted a telephone survey to capture patients’ self-report of vaccination status and to better understand vaccination behavior and attitudes of patients with RA.

This study was conducted at an academic rheumatology clinic from July through September 2013 and was approved by the Northwestern University Institutional Review Board. Structured query language was used to identify all eligible patients from EHR data. Patients were eligible if they had a diagnosis of RA (

International Classification of Diseases, Ninth Revision, Clinical Modification

 codes 714.0-714.9) in the problem list, at least 1 visit to the study clinic in each of the previous 2 years, English listed as their preferred language, and were aged at least 18 years. A total of 1119 patients were identified as eligible based on EHR data. Query output included patient contact information for recruitment. A random number was generated for each patient on the list, and the list was sorted by this number to ensure that a random sample of patients would be recruited.

Our goal was to complete 100 interviews. Assuming a 50% participation rate based on previous work with a similar patient population, we initially selected the first 200 patients for recruitment. Manual chart reviews were done prior to recruitment to confirm that the patients had a diagnosis of RA. After initial chart reviews found that some patients did not have a confirmed diagnosis of RA, we selected an additional 50 patients for possible recruitment. We stopped chart reviews and recruitment of new subjects once we reached 100 completed interviews. Of the 245 manual charts we reviewed, 48 did not have an RA diagnosis, resulting in 197 patients whom we attempted to recruit.

A study interviewer called potential respondents up to 6 times and invited them to participate in a 10-minute structured telephone interview. We varied the time of day for the calls to maximize the chance of contact and to decrease selection bias. A message was left on identified voicemails on the second and fifth attempts asking patients to return the call if they were interested in participating in the study. Patients were offered a $10 gift card for completing the survey.

The survey included published items from the Behavioral Risk Factor Surveillance System survey

 and items developed for this project by the study team. The survey had 22 to 27 items (depending on skip patterns) and addressed the following content areas: a) self-reported receipt of INFVX, PNVX, and HZVX; b) attitudes about these vaccines, including reasons for being unvaccinated, if applicable; and c) provider recommendations about these vaccines. Additionally, if participants reported receipt of a vaccination, we asked for details on the type of setting/place they had received each vaccine. If a participant reported ever receiving an influenza vaccination, we asked if they had been vaccinated during the previous year’s flu season (2012-2013). As HZVX is contraindicated for some immunosuppressed patients, we included an item inquiring whether a provider had ever recommended against receipt of this vaccine. Participants responded to several items regarding attitudes toward vaccinations on a 4-point Likert scale. Open-ended questions were asked to elicit participants’ reasons for not being vaccinated. Interviewers coded responses to open-ended items about reasons for not being vaccinated into the following predefined categories if appropriate: a) no one ever recommended vaccine, b) did not think they needed vaccine, and c) do not like vaccinations/do not think safe. Responses that could not be coded into the above categories were reviewed by 2 authors (DB and DS) and an additional category—d) MD/pharmacist recommended against vaccination&mdash;was added. If more than 1 reason was mentioned, multiple responses were recorded. Demographic items were asked at the conclusion of the survey.

We reviewed EHRs of consented participants who completed the telephone interview to identify whether they were receiving an immunosuppressive drug for treatment of RA, the presence of comorbid conditions, and documented immunizations (administered in the clinic and/or historical vaccinations). We classified patients as receiving immunosuppressive medications if the current medication list included any biologic agents, corticosteroids, methotrexate, leflunomide, or azathioprine; hydroxychloroquine and sulfasalazine were not considered immunosuppressive agents. Comorbid conditions were identified from the participants’ problem list and included cancer, chronic pulmonary disease, coronary artery disease, chronic heart failure, peripheral vascular disease, severe chronic liver disease, diabetes with end organ damage, renal failure, and dementia.

The survey response rate was calculated using guidelines from the American Association for Public Opinion Research.

 Survey responses were summarized using descriptive statistics. We calculated kappa statistics to analyze the agreement between self-reported vaccination and vaccination documentation in EHR data; for these analyses, if a respondent said he or she did not know whether they had been vaccinated, this was categorized as a “no” response.

Of the 197 patients we attempted to recruit, 13 had no valid telephone number, 19 refused participation, 63 were not reached after 6 attempts, and 102 patients enrolled in the study and completed the interview (55.4% response rate). Most participants were female (85.3%) and the mean age was 57.8 years (

). A total of 91.2% of participants were prescribed immunosuppressive therapy based on chart review; 85.2% of participants reported they were taking immunosuppressive therapy.

Most participants (90.2%) reported that they had received INFVX at least once; 79.4% reported receiving INFVX during the 2012-2013 season (

). Of the 81 participants who self-reported INFVX in the last year, only 28 (34.6%) had INFVX recorded in the EHR. This low rate of INFVX documentation in the EHR is consistent with the variety of settings in which patients reported receiving INFVX (

). Of the 21 patients who said they had not received INFVX in the previous year (or were unsure), only 1 (5%) had an INFVX documented in the EHR.

Approximately half of participants (53.9%) reported that they had received PNVX at least once. Only 32 of the 55 (58.2%) who reported getting PNVX had a PNVX documented in the EHR; conversely, of the 47 who said they had never received PNVX, 9 (19.1%) had a PNVX documented in the EHR. If either self-reported or EHR-documented PNVX were included, the rate of receiving 1 or more PNVX was 62.7%.

Only 8 patients (7.8%) reported receiving HZVX; of these, only 1 (12.5%) was documented in the EHR. Of the 94 who said they had not received HZVX, 8 (8.5%) had a documented HZVX in the EHR. Among patients who reported HZVX, 87.5% said they received it at a doctor’s office.

When participants were asked, “How important do you think it is to get vaccinations to prevent infections?” 16.3% said it was somewhat important and 65.3% said it was very important to get vaccinations. However, 15.3% thought vaccines were not important at all. Of note, of the 18 participants who did not feel it was important to be vaccinated, all but 1 were taking immunosuppressive medications at the time of the survey, including 9 taking tumor necrosis factor inhibitors and 12 taking methotrexate. When participants were asked how much they worried about getting infections because of the medications they take for RA, 20% responded “somewhat” and 21% responded “a great deal.”

Participants’ reasons for not being vaccinated varied by type of vaccine. The majority of participants who had not received PNVX or HZVX reported that it had never been recommended to them (

). Patient dislike or distrust of vaccination was more common for INVFX than for PNVX or HZVX. Of the 10 patients who had never been vaccinated for influenza, 7 stated that this was because they did not like or trust the vaccine.

Patient Perceptions of Provider Communication Regarding Vaccines

Three-fourths of respondents said that a physician had told them they had an increased risk of infection (

); however, only 64% recalled being told the importance of vaccines by their providers. Communication about vaccinations also varied by vaccination type (Table 5). Nearly all (96.1%) participants reported that their provider had recommended INFVX, and only 2 patients reported that one of their providers had recommended against INFVX. Only 60.8% said they were told to get PNVX, and only 16.7% reported that HZVX had been recommended (Table 5). Thirteen participants (12.8%) reported they were told not to get HZVX; of these, only 1 patient was eligible for this vaccine (ie, not on biologic therapy).

Based on participants’ self-report, we found a high vaccination rate for influenza during the previous season (79.4%), a moderate rate of any previous vaccination for pneumococcus (53.9%), and a very low rate of any previous vaccination for herpes zoster (7.8%). If we assume that all self-reports are accurate and we include vaccinations recorded in the EHR that were not reported by patients, the vaccination rates were approximately 8% to 9% higher for PNVX and HZVX (Table 2).

A variety of factors likely contribute to suboptimal vaccination rates for patients with RA; lack of physician recommendation, as recalled by the patient, appears to be the most important. Only three-fourths of patients said their doctors had discussed their increased risk of infection, and only two-thirds said their doctors discussed the importance of vaccinations (Table 5). Although almost all participants reported they had been told to be vaccinated for INFVX, less than two-thirds had been advised to get PNVX and only 1 in 6 had been advised to have HZVX. These rates of recommendations correspond well to the self-reported vaccination rates; when patients who said they had not received a PNVX or HZVX were asked why, the most common response was that it was not recommended by their provider.

The reasons why physicians fail to recommend vaccinations for immunocompromised patients are not fully understood. Some failures may occur because patients with RA are often cared for by both a rheumatologist and a primary care physician. The rheumatologist may see preventive care as the responsibility of the primary care physician, but primary care physicians may not be knowledgeable about guidelines for vaccination of immunosuppressed patients or may feel this is the responsibility of the rheumatologist prescribing the immunosuppressants.

A 2009 German study found that only 65% of patients with RA were vaccinated against influenza despite 95% patient awareness of the vaccine’s availability. This was thought to be due to the fact that in Germany, the rheumatologist “delegates” vaccinations to the general practitioner.

Similarly, a 2009 study in Ireland found that approximately half of rheumatologists did not advise influenza vaccination due to insufficient time during the visit, and they also felt vaccination was the responsibility of the general practitioner.

 However, it is not certain whether these European findings are generalizable to the United States. General practitioners in European countries tend to have a more primary role in patients’ care and serve as gatekeepers to specialty services.

 In contrast, in the United States, rheumatologists often share the responsibility of primary care for their patients’ overall health (eg, monitoring cardiovascular risk, blood glucose, malignancy screening), recognizing that many comorbid conditions have a higher prevalence with autoimmune disease. It may be appropriate for specialty societies to address shared responsibility for vaccination and to issue guidance statements to help avoid situations in which responsibility is not clear.

Another factor contributing to lack of recommendation for vaccination is the lack of adequate reminder systems. Patients with RA have complex healthcare needs, and preventive care may be a low priority for providers when they address acute management issues during busy clinic visits. Similarly, physicians have insufficient time to address preventive needs.

 Reminder systems, especially those in EHRs with linked order sets to facilitate prescribing, can help overcome these barriers. Point-of-care paper reminder forms for rheumatologists improved PNVX rates from 67.6% to 80% in patients receiving immunosuppressive medications.

 An EHR alert significantly increased use of both INFVX and PNVX, as well as documentation rates in rheumatology patients taking immunosuppressants.

 Efficacy was better if the process was nurse-driven, which speaks to the potential for team-based care to improve vaccinations.

 The value of such team-based intervention led Kaiser Permanente Southern California to adopt a “Complete Care” program beginning in 2005, which incorporated proactive team-based care, supplemental tactics, and health information technology in order to increase quality of care for patients with chronic illnesses.

Articles

Patients with rheumatoid arthritis self-report a moderate rate of any previous pneumococcal vaccination (54%) and a very low rate of herpes zoster vaccination (8%).

Understanding Vaccination Rates and Attitudes Among Patients With Rheumatoid Arthritis

 To compare quality, utilization, and cost outcomes for patients with selected chronic illnesses at a patient-centered medical home (PCMH) prototype site with outcomes for patients with the same chronic illnesses at 19 nonintervention control sites.

Nonequivalent pretest-posttest control group design.

 PCMH redesign results were investigated for patients with preexisting diabetes, hypertension, and/or coronary heart disease. Data from automated databases were collected for eligible enrollees in an integrated healthcare delivery system. Multivariable regression models tested for adjusted differences between PCMH patients and controls during the baseline and follow-up periods. Dependent measures under study included clinical processes and, outcomes, monthly healthcare utilization, and costs.

 Compared with controls over 2 years, patients at the PCMH prototype clinic had slightly better clinical outcome control in coronary heart disease (2.20 mg/dL lower mean low-density lipoprotein cholesterol; 

 <.001). PCMH patients changed their patterns of primary care utilization, as reflected by 86% more secure electronic message contacts (

 = .003), and 6% fewer in-person primary care visits (

 <.001). PCMH patients had 21% fewer ambulatory care—sensitive hospitalizations 

 <.001) and 7% fewer total inpatient admissions (

 = .002) than controls. During the 2-year redesign, we observed 17% lower inpatient costs (

 <.001) and 7% lower total healthcare costs (

 <.001) among patients at the PCMH prototype clinic.

 A clinic-level population-based PCMH redesign can decrease downstream utilization and reduce total healthcare costs in a subpopulation of patients with common chronic illnesses.

This study investigated healthcare quality, utilization, and costs among patients with common chronic illnesses in a clinic-level, population-based patient-centered medical home (PCMH) prototype redesign durin 2007 and 2008. Multivariable regression models tested for adjusted differences between PCMH patients and controls during baseline and follow-up periods.

		PCMH patients had 7% lower total healthcare costs during the 2-year redesign, largely driven by lower inpatient utilization and costs.

		PCMH patients with coronary heart disease had slightly better low-density lipoprotein cholesterol control.

		Findings demonstrate the capacity of a population-based PCMH redesign to achieve desirable outcomes in a chronically ill patient subpopulation.

Many stakeholders in American healthcare have embraced the patient-centered medical home (PCMH) in recent years. A variety of small and large practices

are implementing pilots and demonstration projects, with financial and operational support from payers

 Although each medical home initiative reflects a unique blend of clinicians, patients, practice infrastructures, and payment mechanisms, all PCMH interventions have the goal of providing patients with a continuous source of whole-person primary care.

Most PCMH interventions emphasize mechanisms to improve care delivery for persons with chronic illness. Chronically ill patients have long been hypothesized to benefit from PCMH elements such as teambased care, productive patient-provider relationships, clinical information  technology use, and delivery system design.

 The chronic care model has been incorporated in PCMH interventions

 and PCMH interventions have disproportionately targeted chronically ill patients

 or elderly patients with high chronic illness burdens.

Despite these links between the medical home and chronic illnes care, the evidence base contains few, if any, rigorous evaluations of PCMH effects on the quality, utilization, and costs of care in patients with chronic illnesses. We address this gap by reporting findings of a 2007 to 2008 prototype PCMH redesign

 among patients with at least 1 of 3 common chronic illnesses in which the majority of care is typically delivered in the primary care setting: diabetes, hypertension, and coronary heart disease (CHD). Our objective in conducting this study was to investigate differences in quality, utilization, and costs of care between chronically ill patients at the PCMH site and comparable patients at 19 nonintervention control sites in the same healthcare system.

We assessed the impact of a PCMH redesign implemented at 1 clinic within Group Health, an integrated health plan and care delivery system in Washington State. The PCMH prototype clinic is located in metropolitan Seattle and is one of 20 clinics Group Health owns and operates in Washington’s Puget Sound region. The clinic was chosen as the PCMH prototype because of the stability of its leadership and its history of successfully implementing change. Group Health pursued the PCMH redesign after a series of reforms in financing and primary care operations yielded mixed results.

 Although the earlier reforms achieved their primary objectives of increasing patient access and satisfaction with care and reducing total costs, discouraging trends (eg, increased emergency department [ED] costs, decreased job satisfaction among primary care physicians) were also observed.

A comprehensive list of design principles and change components in the PCMH redesign is presented elsewhere,

but we describe selected key elements here. In the prototype clinic, increased primary care staffing supported reductions in physicians’ patient panels from an average of 2327 patients to 1800 patients, physicians were paired in dyads with medical assistants, and standard in-person primary care office visits were lengthened from 20 to 30 minutes. “Virtual medicine” contacts—secure electronic messaging and telephone encounters&mdash;were emphasized by encouraging patients to register for a secure online patient portal and by rerouting patients’ calls to an organizational consulting nurse service to primary care teams during normal clinic operating hours. Some PCMH components explicitly targeted chronically ill patients,

 such as creation of collaborative care plans and provider outreach (by phone or secure message) to manage monitoring tests.

Prior analyses compared 2-year outcomes for patients at the PCMH prototype clinic with those for patients at other Group Health clinics in western Washington State.

 In both the full practice and the practice’s elderly subpopulation, PCMH patients had fewer ambulatory care—sensitive hospital admissions (13% full practice, 18% elderly) and fewer combined ED and urgent care visits (29% full practice, 21% elderly). Six percent fewer all-cause hospitalizations and accompanying lower inpatient costs ($14 per month) were also observed in the full practice.

This study used a nonequivalent pretest-posttest control group design,

including baseline data from 2006 and followup data from 2007 and 2008. We used automated Group Health databases to identify adults with diabetes mellitus (types 1 and 2), hypertension, or CHD. These data sources contain diagnoses, procedures, and pharmacy data for care obtained at Group Health and at sites where providers deliver care to Group Health patients on a contracted basis; laboratory results and clinical encounter data are only available for care provided at Group Health. The accuracy and completeness of these data sources have been extensively validated.

 Group Health’s institutional review board approved all study protocols.

Patients in the final study population were aged 18 to 85 years, received care at 1 of 20 Group Health clinics in western Washington State, had at least 6 months of enrollment during 2006, and had 3 or more months of enrollment in both 2007 and 2008. We also required enrollment during December 2006, which facilitated collection of baseline case mix variables.

 To account for clinic-level factors and ensure comparability across study groups, we excluded patients who switched enrollment between clinics on a year-to-year basis. We excluded patients with dementia at baseline and women who gave birth during the study, as much of their healthcare use was presumably attributable to factors external to the PCMH redesign.

Patients at both the PCMH clinic and other clinics were only included in the final study population if they had 1 or more of the 3 included chronic illnesses. We identified patients with preexisting diabetes, hypertension, and CHD using case definitions designed to achieve high specificity and high positive predictive value.

 This approach utilized patterns of diagnoses, procedures, laboratory values, and pharmacy fills to minimize erroneous inclusion of “false positive” patients with unconfirmed chronic illness. Case definitions are listed in 

We collected data on disease-specific quality of care in the years 2006 and 2008. Laboratory results provided glycated hemoglobin (A1C) levels for patients with diabetes and low-density lipoprotein (LDL) cholesterol levels for patients with CHD. Systolic and diastolic blood pressure readings for patients with hypertension were acquired from electronic encounter data. If patients had clinical outcome data collected more than once in an individual year, we only used the last recorded value from that year.

Laboratory and blood pressure data were converted into disease-specific dependent variables for 3 types of quality measures: clinical processes, clinical outcome benchmarks, and mean clinical outcomes. We created binary measures of clinical process performance based on whether laboratory data on A1C and LDL were collected annually. Outcome benchmarks were assessed by binary measures of A1C below 9.0% among patients with diabetes, blood pressure below 140/90 mm Hg among patients with hypertension, and LDL below 100 mg/dL among patients with CHD. Continuous A1C, systolic blood pressure, and LDL cholesterol results provided mean clinical outcome measures for each chronic illness.

Utilization and cost data were collected for the 2006 baseline year and the 2-year PCMH redesign. Group Health’s automated systems assigned patient costs on a monthly basis, reporting actual costs from the general ledger. Overhead costs (eg, additional staffing costs during the PCMH redesign) were fully allocated to patient care departments. Cost and in-person utilization data were collected for primary care, specialty care, total inpatient admissions, and combined ED and urgent care. We also collected data on  ambulatory care—sensitive inpatient utilization

 and total healthcare costs.  Clinical databases provided data on patients’ use of secure message threads, telephone encounters, and calls to the consulting nurse service. We accounted for changes in internal cost accounting at Group Health during fall 2008 by truncating collection of cost and in-person utilization data at 21 months, which ensured consistency in these variables over time. Period-specific utilization and costs were converted to monthly rates based on  patients’ number of days of enrollment at Group Health during the baseline and PCMH redesign.

We collected data on patients’ age, sex, and case mix variables from Johns Hopkins Adjusted Clinical Groups (ACG) System software.

Aggregated Diagnosis Groups variables from December 2006 provided a measure of patients’ morbidity burden during the 2006 baseline year. The 32 Aggregated Diagnosis Groups variables classify International Classification of Diseases, Ninth Revision diagnoses into clinically cogent morbidity clusters based on duration, severity, diagnostic certainty, etiology, and expected need for specialty care, and have been extensively used for case mix ascertainment and adjustment in primary care populations.

Multivariable regression models assessed differences between PCMH patients and controls during the 2006 baseline year and the 2007 to 2008 follow-up periods. Each model tested for the effect of the PCMH by including a patient-level indicator of empanelment at the PCMH prototype clinic as the independent variable.

Quality-of-care analyses used data from the 2006 baseline year and 2008 to investigate differences between PCMH patients and controls with respect to condition-specific processes and outcomes. All patients with diabetes and CHD were included in the clinical process analysis; the clinical outcomes analysis for each chronic illness was restricted to patients who had disease-specific outcome data collected in both 2006 and 2008. We used Poisson regression models (Poisson distribution, log link) to investigate clinical processes and outcome benchmarks in the 2006 baseline year and the 2008 follow-up year, incorporating robust variance estimates to obtain relative risks.

We used linear regression models to investigate differences in mean clinical outcomes in 2006 and 2008. Each condition-specific regression model was restricted to patients with the targeted chronic illness, and adjusted for age and sex. Regression models estimating 2008 results additionally adjusted for 2006 baseline results.

Our analysis of healthcare use investigated rates of monthly in-person contacts and virtual medicine use during the 2006 baseline, the redesign’s first year (2007), and the full observation period (in-person utilization analysis was truncated at 21 months). We used Poisson regression models to estimate  utilization rates, setting the scale parameter to the deviancedivided by the residual degrees of freedom to preclude violations of model assumptions of over- or under-dispersion.

 Regression models estimating baseline  healthcare use adjusted for age and sex. Regression models estimating use during the PCMH redesign adjusted for age, sex, Aggregated Diagnosis Groups count, and binary indicators for each of the 3 chronic illnesses.

We used an algorithm recommended by Manning and Mullahy32 to select models for cost outcomes that would provide unbiased estimates of covariate effects in right-skewed monthly cost data. This approach led us to select a generalized linear regression model with log link and gamma distribution and to report results in proportional—rather than dollar-based&mdash;cost differences. Regression models estimating baseline year costs  were adjusted for age and sex. Regression models estimating PCMH redesign period costs were adjusted for age, sex, and log-transformed baseline costs for each dependent cost variable (eg, inpatient cost models adjusted for log-transformed baseline inpatient costs).

All regression models investigating quality, utilization, and cost in the chronically ill study population were estimated using generalized estimating equations with independent working correlations and robust sandwich variance estimates.

 This approach controlled for clustering at the level of patients’  paneled primary care clinics and is robust to misspecification of within-cluster correlation. Analyses were conducted using SAS software, version 9.2 (SAS Institute Inc, Cary, North Carolina). We used an alpha threshold of 

.05 to determine the statistical significance of findings.

We conducted multiple sensitivity analyses to assess whether cost and utilization results were robust across analytic approaches. When assessing stability of 12- and 21-month cost and utilization results, 1 sensitivity analysis included patients with asthma and chronic obstructive pulmonary disease (COPD). These patients were initially targeted for inclusion but were excluded from the final study population due to low prevalence at the PCMH clinic and unavailability of diseasespecific quality outcome data. A utilization-only sensitivity analysis by using DxCG risk scores mimicked the risk adjustment approach of prior PCMH prototype analyses.

 An additional sensitivity analysis used a difference-in-difference approach, adjusting for age and sex, to assess potential variability of utilization and cost findings for the full follow-up period.

The final study population included 37,938 adults with diabetes, hypertension, and/or CHD, with 1181 patients paneled to the PCMH prototype clinic and 36,757 patients paneled to other clinics (

). On average, PCMH patients were 3.1 years older than patients at other clinics (mean age 65.0 vs 61.9 years; 

 <.001) and more likely to be female (54% vs 51%; 

= .04). Patients at the PCMH clinic were less likely to have comorbid combinations of diabetes, hypertension, and CHD (13% vs 21%; 

 <.001), but had higher counts of Aggregated Diagnosis Groups case mix variables (mean 7.2 vs 6.9;

 = .01), reflecting a greater prevalence of health conditions other than the 3 conditions targeted in this study. Among patients at both the PCMH prototype clinic and controlclinics, 93% were enrolled at Group Health for the entire 2-year follow-up period.

Quality-of-care analysis revealed differences between the PCMH clinic and control clinics (

). During the 2006 baseline year, patients with diabetes at the PCMH clinic were 4% more likely to have A1C under 9.0% (

 <.001) and their mean A1C was 0.20% lower than that of the controls (

 <.001). These differences among patients with diabetes largely persisted through the PCMH redesign. Although we observed no significant differences in LDL control among patients with CHD during baseline, in 2008 patients at the PCMH clinic were 11% more likely to have LDL below 100 mg/dL, (

 <.001), and their mean LDL cholesterol was 2.20 mg/dL lower than that of the controls (

 <.001). Patients with hypertension at the PCMH clinic were 5% more likely than controls to have blood pressure below 140/90 mm Hgduring baseline (

= .01), but we observed no differences in blood pressure control in 2008.

During the pre-PCMH baseline year, patients at the PCMH clinic had higher rates of specialty care visits (10%, 

 <.001), ambulatory care—sensitive hospitalizations (31%,

 <.001), and total inpatient admissions (22%, 

). Baseline period differences  in inpatient care were reversed during the first yearof PCMH redesign—patients at the PCMH clinic had 23% fewer ambulatory care—sensitive hospitalizations (

<.001) and 4% fewer total inpatient admissions (

 = .05) than controls—and these differences were sustained over 21 months. Patients at the PCMH clinic had fewer ED and urgent care contacts than controls during the baseline period (18%, 

 = .005); this difference widened to 29% over the first year of the PCMH redesign (

 <.001). Patients at the PCMH prototype clinic had 6% fewer in-person primary care contacts than controls (

Patients at the PCMH prototype clinic had more virtual medicine contacts than controls during the pre-PCMH baseline year (

.05). During the first 12 months of the redesign, PCMH patients had 94% more secure electronic message threads (

<.001) and 17% more telephone encounters than controls (

 <.001). These differences were attenuated slightly in 24-month results (

003). The PCMH patients had 18% and 22% fewer telephone calls to the consulting nurse service over 12 and 24 months, respectively (

 <.001 for both). This finding is consistent with the redesign’s goal of decreasing nonemergent consulting nurse calls by redirecting calls to the primary care team during clinic operating hours.

Baseline cost findings were similar to observed baseline utilization results, including higher specialty care costs (32%, 

 <.001), lower ED and urgent care costs (16%, 

 <.001),and higher total inpatient costs (33%, 

<.001) among patients at the PCMH clinic (

). Despite nonsignificant differences in specialty utilization during the PCMH redesign, specialty care costs were 9% higher among PCMH patients over both 12 months (

 = .004) and 21 months (P = .02). The PCMH patients had 6% lower ED and urgent care costs over 21 months (

= .01), 16% lower inpatient costs over 12 months (

 <.001), and 17% lower inpatient costs over 21 months (

 <.001). Total per member per month costs were 19% higher among PCMH patients during the baseline year (

In sensitivity analyses using alternative inclusion criteria and case mix variables, findings for virtual medicine use and 21-month in-person utilization and costs were robust across analytic approaches; selected 12-month findings varied by approach. When adjusting for DxCG variables or including patients with asthma and COPD (61 additional PCMH patients and 2340 additional  controls) in 12-month analyses, we observed 5% greater specialty utilization among PCMH patients (

 = .02 including asthma and COPD) and no significant differences in all-cause inpatient admissions. Including patients with asthma and COPD also led to the observation of 7% higher primary care costs among PCMH patients at 12 months (

The sensitivity analysis using a difference-in-difference approach estimated more dramatic savings among PCMH patients for inpatient costs and total healthcare costs. Appendix B presents complete difference-in-difference results and accompanying comments.

We observed numerous associations between a PCMH redesign and the quality, utilization, and costs of care among patients with diabetes, hypertension, and CHD. Patients at the PCMH clinic changed their primary care utilization patterns over the 2-year redesign, as reflected by use of more phone encounters and secure message threads, but fewer in-person primary care visits, than controls. Despite these changes, we observed no group-level differences in primary care costs, presumably due to the longer length of in-person visits at the PCMH site and costing systems’ allocation of PCMH implementation costs to the prototype clinic. The PCMH patients had 9% higher specialty care costs during the redesign, which is smaller than the 32% greater specialty costs during the baseline year, but is consistent with prior findings.

Over 21 months, we observed reduced utilization of downstream care at the PCMH site, including less ED and urgent care, fewer ambulatory care—sensitive hospitalizations, and fewer total inpatient admissions. The PCMH patients had 7% lower total healthcare  costs during the PCMH redesign,largely driven by the reversal of baseline year trends in inpatient use and costs. When applied to the $697 unadjusted total monthly healthcare costs in this study population during the 2006 baseline year, the observed 7% cost difference translates to approximately $49 savings per month.

Improved clinical outcome control among patients with CHD at the prototype clinic provides evidence of the PCMH’s ability to globally reduce costs in this population while improving quality. Although differences in LDL cholesterol were statistically significant, they were small in magnitude, which is consistent with modest quality improvements observed in other PCMH transformations.

Our 2-year results on setting-specific utilization are comparable to those in other patient populations at the PCMH prototype clinic.

 We observed slightly greater differences between PCMH patients and controls for emergency and urgent care contacts, ambulatory care—sensitive hospitalizations, and total inpatient admissions. Given similar findings across analyses (eg, 7% fewer inpatient admissions among the chronically ill patients and 6% fewer among the full clinic population), we do not believe the impacts of the PCMH redesign were confined to patients with chronic illness. However, utilization impacts in chronilogically ill patients wereconsistently larger than those in prior analyses, which probably contributed to the statistically significant lower total healthcare costs observed here, but not for all patients.

This study differed from prior medical home evaluations in important ways. Unlike several PCMH interventions in the research literature,

 Group Health’s PCMH prototype redesign did not embed care managers in a primary care practice, nor was the intervention confined to a subgroup of high-risk or chronically ill patients. It was instead implemented in the prototype clinic’s full patient population, incorporating fundamental tenets of primary care

 to support productive, longitudinal relationships between physician-led primary care teams and patients. Consistent with recommendations of a recent white paper commissioned by the Agency for Healthcare Research and Quality,

 our study investigated PCMH effects among chronically ill patients and adjusted for clinic-level clustering.

Our encouraging findings were presumably attributable to an alignment of structures and processes at the PCMH clinic that allowed for effective provision of whole-person care. The  PCMH components that may have facilitated the productivepatient-provider interactions envisioned by the chronic care model

 include enhanced care team staffing, pairing longer office visits with promotion of increased virtual medicine use, and effective outreach for patients’ chronic and acute needs.

This study investigated healthcare quality, utilization, and costs among patients with common chronic illnesses in a patient-centered medical home prototype redesign.

David T. Liss, PhD

Articles

Understanding Vaccination Rates and Attitudes Among Patients With Rheumatoid Arthritis

Outcomes Among Chronically Ill Adults in a Medical Home Prototype