With growing pressure to improve the quality and coordination of care, physicians feel a need to streamline their relationships with other practitioners around shared care for patients. Some physicians have developed written agreements that articulate the respective responsibilities of 2 or more parties for coordination of patient care, ie, care coordination agreements (CCAs).

To describe how CCAs are formed and explore facilitators and barriers to adoption  of effective CCAs, the extent to which CCAs may be replicable in different market contexts, and the implications for policies and programs that aim to improve the coordination of care.

Qualitative study of primary care physicians participating in CCAs and representatives of their specialist, hospital, or communitybased partners.

Semi-structured interviews with participating providers and national thought leaders in care coordination were reviewed to develop key themes.

Agreements that address referral and access processes were considered useful by all practices that had implemented them. Practices that implemented agreements including guidance on shared management of specific clinical conditions (comanagement) also found them useful. CCAs were most successful in settings where both parties to the agreement already had stable communication pathways (such as an electronic health record [EHR], designated staff) and strong working relationships.

Policy changes (such as shifts in reimbursement to favor collaborative care or clarification of laws governing such collaborations) can help to support the development and implementation of CCAs, and can address factors that currently make some markets less supportive of coordination.

Primary care providers seeking to improve their patients’ access to and coordinate care with specialists, hospitals, and community-based organizations may struggle to build and maintain relationships with these entities. Care coordination agreements are written agreements that delineate the way primary care practices will share care responsibilities with these partners.

		Practices that have piloted this approach report improved access and quality of care, but note several pitfalls that can limit their utility.

		Changes in policy and care reimbursement can encourage further development of these agreements.

Physicians face significant challenges in coordinating care. Patients (particularly those with chronic illnesses) often see many providers; 1 study found that the typical fee-for-service Medicare beneficiary sees 7 different physicians in 4 different practices during 1 year.

 Poor coordination may result in patients receiving conflicting advice from different providers (including conflicting medication prescriptions), and in lack of follow-up for new diagnoses, tests, or procedures posing risks to patients’ health.

 The burden of care coordination may be particularly high for primary care providers (PCPs), who are charged with serving as the hub for communication and decision making between patients and all of their providers.

While there is no single recipe for effective coordination, one promising strategy is the use of “service agreements” or “care coordination agreements.” Care coordination agreements (CCAs), written agreements that articulate the respective responsibility of 2 or more parties for ensuring coordination of patient care, can be helpful tools that allow PCPs to define the level of care expected from themselves and other providers, and to set out specific tasks for which each partner will be responsible. Similar care coordination efforts organized by insurers have been associated with decreased costs and, in some cases, increased quality of care.

 The American College of Physicians has endorsed further exploration of these agreements, particularly in the context of the patientcentered medical home,

 and has developed a set of model agreements.

The National Committee for Quality Assurance’s 2011 standards and guidelines for the Patient-Centered Medical Home allow the development of service agreements between PCPs and specialists to count toward accreditation.

CCAs have been adopted by a diverse group of primary care practices and in varied market contexts. A limited literature

suggests that CCAs can improve care coordination in certain tasks. There is also anecdotal evidence to suggest that CCAs are more common than they appear to be in the formal literature.

However, little research has been done to describe the circumstances under which CCAs are most likely to develop, to discuss common features, and to weigh the implications of having these agreements in place.

Care coordination agreement cases are defined for the purpose of this study as dyads of a PCP and a partner (specialty practice, hospital, long-term care provider, urgent or emergent care provider, or community-based organization) that have agreed for at least 6 months to coordinate the care of shared patients. Further, CCAs must involve, if not focus on, delineating responsibilities of each party in the agreement so as to better coordinate patient care between a PCP and the other provider. The agreements should be in the form of a written document, or else have generated written supporting materials such as standardized referral forms, guidelines, protocols, or electronic medical record (EMR) support.

Under these criteria, practices were identified across the United States through iterative networking via contacts and e-mail listservs for Patient-Centered Medical Home initiatives as well as national primary care, specialty care, urgent care, and pharmacy societies. We purposively selected practices/organizations to include a range of characteristics in terms of size (small, medium, and large practices), ownership type, geography, and rural/urban and suburban settings. During a screening call, we identified practices/organizations that had a CCA supported by written agreements that the partners were willing to make available to the research team.

In all, 37 telephone interviews were conducted from July 2010 to April 2011 with 6 national experts who have published extensively in the peer-reviewed literature on care coordination and related issues or are leaders of national organizations at the forefront of policy related to care coordination, and 31 healthcare providers representing 15 partnerships. (One PCP was interviewed regarding 2 separate partnerships.) Respondents included independent practitioners, hospital-employed physicians, and physicians working in not-for-profit, academic, safety-net, and Veterans Affairs settings. Urban, suburban, and rural respondents were represented. No participants refused to be interviewed.

Notes were transcribed verbatim and jointly reviewed by both the lead interviewer and the research assistant for quality and validation purposes. The interview responses were coded and analyzed using Atlas.ti, a qualitative data management and analysis software tool.

Participants were required to submit a copy of each CCA or its supporting material to analyze its content and verify respondents’ assertions. A database of selected CCA agreements is available upon request to the authors.

We identified 15 CCAs. Five of the partnerships were with specialist practices, 6 were with hospitals, 5 involved community-based organizations such as behavioral health providers, and 1 involved an emergency care provider. (In some cases, agreements involved more than 2 participants.) Six agreements exclusively involved practices and organizations working with children; the remainder involved practices treating adults or all ages. Six were staffed by 5 or fewer physicians, and an additional 5 were staffed by 6 to 10 physicians; the remainder were larger practices. Through these CCAs, providers divide up the work of patient care, using some combination of broad strokes aimed at defining the relationship between 2 practices and specific details that help to standardize joint management of particular conditions. These agreements are generally developed at the practice or Independent Practice Association level rather than by individual physicians within practices. The following section describes attributes of CCAs in greater depth (see 

for the frequencies with which various elements appear in CCAs).

Based on themes raised in respondents’ descriptions of their agreements, we identified 2 principal types of CCA. The first, which at times has been called a 

 includes broad agreements as to how providers should generally conduct referrals and transitions of care. An example of one such agreement between primary care providers and specialists requires participants to “define responsibilities between PCP, specialist, and patient” and “maintain competency and skills within scope of work and standard of care.” The second type are 

, which delineate specific understandings regarding division of tasks for comanagement of particular conditions (for example, a detailed algorithm outlining the process by which children with pediatric voiding dysfunction will be first evaluated by their pediatrician and then, if necessary, referred to a pediatric urologist). Many CCAs include elements of both types (for example, a master service agreement setting general expectations for interactions between participants as well as specific comanagement agreements for conditions of particular interest). (See Table 1, listing CCA characteristics among the practices evaluated for this study.)

Coordination tasks most commonly addressed by existing CCAs include ensuring timely access for patients and standardizing the content in information transfer, as well as ensuring timeliness of information transfer and clear division of clinical comanagement responsibilities.

Nearly every agreement addresses the mechanics of how referrals are requested and conducted. Respondents reported that limited access to referrals was often the strongest motivation to initiate the agreement in the first place. In general, the more standardized the referral process described in a CCA, the more consistently partners tended to apply the agreement and the more partner practices praised their utility.

Elements detailing the shared management of specific conditions were found in roughly half of the agreements we reviewed, and were described by all respondents who had developed them as a useful feature. For example, a comanagement agreement between a PCP and a pharmacist stipulates that for diabetic patients the pharmacist can order specific tests under the physician’s license and can adjust medications according to an agreed-upon protocol based on the test results.

		Specifications for proper referral preparation for selected conditions, including transfer of records, workup, and education of the patient prior to specialist evaluation

		Continuing medical education for PCPs on comanagement of certain conditions

Articles

WEB EXCLUSIVE

Primary care providers have developed standing agreements with other healthcare providers and community-based organizations to coordinate care. Early experiences with these agreements are discussed.

Care Coordination Agreements: Barriers, Facilitators, and Lessons Learned

To examine the correlation of preexisting illnesses and preoperative medical expenditures with postoperative major adverse outcomes among geriatric surgical patients.

Retrospective cohort study using claims from Taiwan’s National Health Insurance Research Database.

65 years receiving inpatient surgeries during 2004 to 2007 under universal healthcare coverage were included. Surgical patients aged 55 to 64 years were the reference group. Risk-adjusted 30-day postoperative complication and mortality rates among elderly patients in various age sectors were analyzed and correlated with the preexisting illnesses and preoperative medical expenditures quantitatively.

Among 432,614 elderly surgical patients in specific age sectors and 238,802 controls, the prevalence of preexisting illnesses and the riskadjusted postoperative adverse outcome rates were highly age dependent and illness related. When comparing patients aged 

85 years with patients aged 55 to 64 years, the adjusted odds ratios were 2.74 (95% confidence interval [CI], 2.67-2.82) and 3.56 (95% CI 3.31-3.84) for incidence of major postoperative complications and mortality after major complications, respectively. Numbers of preexisting illnesses increased in an age-dependent pattern and the preoperative 24-month medical expenditures increased incrementally with the numbers of comorbidities. Postoperative major complications, but not mortality rates, were highly correlated with the numbers of comorbidities and increased parallel with preoperative 24-month comorbidity-related medical expenditures, especially in the younger age group.

Adjusting for preexisting covariates, geriatric patients had an age-dependent, illnessrelated, and expenditure-associated pattern of higher postoperative complication and mortality rates. The numbers of comorbidities and preoperative medical expenditures had high predictive value for postoperative adverse outcomes.

Age range, number and types of preexisting illnesses, and preoperative 24-month medical expenditures were used preoperatively to predict postoperative complication and mortality rates for geriatric surgical patients who had universal healthcare coverage from the Taiwan National Health Insurance Program.

		Postoperative major complications, but not mortality rates, were highly correlated with the numbers of comorbidities.

		Complication rates increased parallel with preoperative 24-month comorbidity-related medical expenditures

Acute care for inpatient surgery is an effective indicator for assessing healthcare system quality because of clear postoperative outcomes such as complications and subsequent mortality.

Previous large-scale population studies on surgical care quality (eg, the National Veterans Affairs’ Surgical Quality Improvement Program, the Surgical Care Improvement Project of the Centers for Medicare & Medicaid Services in the United States) mostly addressed specific populations under certain insurance systems, with important variations in follow-up regarding hospital performance and patient care.

Among healthcare beneficiaries, elderly patients scheduled for major in-hospital surgery are at relatively higher risk for complications due to their complex preexisting comorbidities.

 Although multidisciplinary approaches have significantly improved geriatric patient care in recent decades, and several specialized fields have documented associations between surgeries and patient outcomes,

assessment of outcomes specifically showing the overall quality of healthcare under universal coverage is lacking. The increasing need for medical therapies among the elderly also leads to increased medical costs.

 Financial stresses have raised concerns among health policy makers and regulators of healthcare providers regarding costs and benefits for the elderly and their postoperative outcomes. However, studies are still lacking that comprehensively analyze (1) the impact of geriatric patients’ characteristics (ie, preexisting comorbidities) on surgical complications or mortality and (2) the potential role of preoperative medical expenditures in predicting postoperative outcomes for geriatric inpatient surgery.

This was a large-scale, nationwide, and population-based study under a uniform and homogeneous healthcare system to verify the quality of acute care in terms of postoperative complication and mortality rates for elderly patients who underwent in-hospital surgery. We conducted a retrospective cohort analysis of beneficiaries aged 

65 years under the universal coverage of Taiwan’s National Health Insurance Program (NHIP).

 Demographic and functional status, coexisting illnesses, and preoperative medical expenditures among different age groups of elderly patients were analyzed for riskadjusted 30-day postoperative complications and mortality rates after receiving inpatient surgery. After adjustment for patients’ preexisting medical conditions and preoperativemedical expenditures, we validated the correlations of comorbidities and preoperative medical expenditures with postoperative adverse outcomes for elderly surgical patients. These correlations can serve as potential predictors of clinical outcomes and provide future guidance regarding resource allocation when considering healthcare policy reforms.

Medical claims identified from Taiwan’s National Health Insurance Research Database were provided by the Department of Health’s National Health Insurance Bureau.

The data set includes all claims from Taiwan’s NHIP, a single-payer system begun in 1995, which finances care for all citizens and offers them unrestricted access to any healthcare provider.

 Currently NHIP covers more than 22.6 million enrollees, representing 99% of the population.

The bureau reviews all reimbursement claims filed by contracted health organizations and screens the type, volume, quality, and appropriateness of medical services provided.

The research database’s registration files include patient demographics, principal and secondary diagnostic codes, procedure and prescription codes, claim summaries, detailed orders, prescriptions and expenditures for inpatient and ambulatory care, and postoperative 30-day morbidity and mortality data.20 The study contents (eg, diagnoses, procedures, medical utility parameters from the National Health Insurance Research Database) have been fully validated by the National Health Research Institutes and the Bureau of National Health Insurance, Taiwan. Besides the administrative and financial routine validation, the National Health Insurance Research Database was further validated in a recent study, suggesting that the research database is a valid resource for population-based research.

Patients enrolled in this study were those discharged from acute care hospitals between January 1, 2004, and December 31, 2007, with principal/secondary diagnoses and procedures based on International Classification of Diseases, Ninth Revision, Clinical Modification (

) codes. All personal identifiers were decoded before the data were released for research, and the National Health Insurance Research Database committee evaluated and approved the study.

The initial 2004 to 2007 database included information on more than 2 million National Health Insurance beneficiaries who underwent inpatient surgery under anesthesia at contracted medical facilities. We limited the study to patients who had procedures performed under general, spinal, or epidural anesthesia, and excluded surgeries with very low mortality rates such as ophthalmologic, auditory, nasal, and other minor procedures under regional or local anesthesia. Considering all the risk factors (primary variables) among the elderly, age difference and complex comorbidities were the major concerns and contributory factors for the outcomes of the study.

 We selected geriatric patients (by definition aged 

65 years) and categorized them into strata of ages 65 to 74, 75 to 84, and 

85 years. These patients received inpatient surgeries in 2004 to 2007 under universal healthcare coverage from the Taiwan NHIP. Using the inclusion criteria identified in this paragraph, we selected surgical patients aged 55 to 64 years, the closest age group, as the reference group. We then segregated data on patients who received inpatient surgery and were admitted for more than 1 day in order to exclude outpatient surgeries and to take only an individual’s first operation into consideration. In that way, we avoided the potential interference of surgeries derived from the postoperative complications per se if patients underwent more than 1 procedure during this period. As we restricted our analysis to patients aged >55 years, 671,540 patients were identified. Excluding missing data in 124 claims left 671,416 patients. The 432,614 patients aged 

65 years were classified into subgroups of 65 to 74, 75 to 84, and 

85 years. The 238,802 patients aged 55 to 64 years served as the reference group (

Demographic characteristics included sex, preoperative functional status, and 15 other variables that identified additional major preoperative coexisting illnesses (ie, hypertension, diabetes, transient ischemic attack/stroke, chronic obstructive pulmonary disease, myocardial infarction/ischemia, dialysis, congestive heart failure, peripheral vascular disease, acute renal failure, ascites, bleeding disorder, emergent operation or not, and previous use of medications for or management of conditions such as preoperative transfusion of more than 4 units, use of vasopressors/inotropes, and albumin supplementation).

 The coexisting illness variables were identified from all inpatient and outpatient administrative health insurance claims of individual beneficiaries submitted since January 1, 2002, which provided data for 24 months prior to surgical admission date. Other potential variables included the number of medical illnesses and preoperative 24-month medical expenditures (all inpatient and outpatient medical services and medical resource consumption including prescription, laboratory, and radiologic examinations, preoperative evaluation and management, physician fees, etc) for patients enrolled. Individual healthcare expenses were collected and analyzed based on healthcare claims for comorbid conditions within 24 months preoperatively for the index surgery (

Complications and mortality after complications were categorized into 8 major groups, defined by the primary and secondary diagnoses according to the 

 codes assigned at discharge and by subsequent inpatient or outpatient records within 30 days of surgery. Major postoperative complications included pneumonia (

 code 584), acute myocardial infarction (

Distributions of demographic characteristics, preoperative functional status, prevalence of coexisting conditions, observed complication rates, and mortality rate after complications among the 4 age strata were compared using the c2 test. Multiple logistic regression models were used to estimate the odds ratio (OR) and 95% confidence interval (CI) of mortality associated with the postoperative complication status or numbers of coexisting medical conditions in different age groups. All models were adjusted for demographics, functional factors, and preexisting illnesses. In addition, the adjusted ORs of mortality related to different numbers of coexisting medical conditions were evaluated among the different age strata. Quartile values of preoperative medical expenditures were used as a cutoff for the trend test, and the rates of complications or mortality for 4 strata of preoperative medical expenditures were further evaluated after multivariate adjustment. Groups with low preoperative medical expenditures were used as a reference to estimate the adjusted ORs for other groups. All data analyses were performed using SAS System for Windows, version 8.2 (SAS Institute, Inc, Cary, North Carolina). A 2-sided 

 value of less than .05 was considered significant.

The prevalence of the 15 major preexisting medical conditions listed above under “Primary Outcomes” increased with age, as well as preoperative medications such as transfusion of red blood cells and use of vasopressors/inotropes and albumin (

 <.0001). Conditions showing age differences but not age dependence were ascites and bleeding disorders (

 = .0037 and .0127, respectively). Among the illnesses, the incidences of hypertension, myocardial infarction/ischemia, and peripheral vascular disease were age dependent for patients aged 55 to 84 years, while they decreased in patients aged 

85 years an age greater than the population’s average life expectancy (

In the reference group (aged 55-64 years), the overall postoperative complication rate for the 8 major postoperative complications listed above under “Primary Outcomes” was 11.94%, which increased to 18.05%, 26.63%, and 34.12% in the 65 to 74, 75 to 84, and 

85 year age groups, respectively. The risk ratio of complications was 2.86 for the group aged 

85 years versus the reference group. After adjusting for sex, preoperative functional status, and coexisting conditions, the riskadjusted complication rates varied among the study groups, ranging from 15.52% in the reference group to 23.61% in group aged 

85 years, with a risk ratio of 1.52 versus the reference group (

). Although we focused mainly on 8 major postoperative complications, the mortality after these major complications in our cohort accounted for 66.70% of overall deaths from all causes in the database. The crude mortality rate after major complications was 18.10% in the 

85 year age group compared with 5.52% in the reference group, with a risk ratio of 3.28. After adjusting for sex and preexisting illnesses, the risk-adjusted mortality rate after complications increased from 8.64% in the reference group to 10.17% in the group aged 

85 years, with a risk ratio of 1.18. For each variable, as age increased there were significant increases in the number of major postoperative complications such as pneumonia, acute renal failure, pulmonary embolism, septicemia, and stroke. Adjusted ORs ranged from 1.91 to 5.21 when comparing the 2 extreme age groups (

 <.0001). In contrast, the incidences of postoperative bleeding and deep wound infection significantly decreased with age (

Most geriatric surgical patients had at least 1 comorbidity associated with complications and mortality (77.4% vs 57.4% of the reference group; 

). Numbers of medical conditions increased with age. The reference group had the highest percentage of patients without preoperative medical conditions, and the group aged 75 to 84 years had the lowest percentage (42.56% vs 19.18%, 

 <.0001). The incidence of patients with 5 or more preexisting conditions was much greater in those aged >85 years than in the reference group (3.65% vs 1.09%; 

 <.0001). The proportions of patients with 2 or more medical conditions were similar in the groups aged 75 to 84 and 

85 years; the groups aged 55 to 64 and 65 to 74 years had significantly lower proportions of those patients (

 <.0001) (Figure 2A). Considering the number of medical conditions and postoperative outcomes in geriatric surgical patients, the complication and mortality rates increased with an incremental number of coexisting conditions among all age groups, and the incidence was highest in patients with 5 or more comorbidities (

). After adjusting for preoperative status, the OR of complication rates in patients with 5 or more medical conditions compared with patients who had none was 7.23 (95% CI 6.63-7.89) in the reference group and 4.04 (95% CI 3.56-4.59) in patients aged 

The correlation between mortality rates after major complications and numbers of coexisting conditions displayed a different and relatively unaffected pattern in elderly surgical patients when they had fewer than 3 comorbidities. The adjusted OR was significant at 1.65 (95% CI 1.34-2.03) and 1.97 (95% CI 1.53-2.53) in the reference group and at 1.44 (95% CI 1.19-1.75) and 1.79 (95% CI 1.43-2.25) in the group aged 

85 years when comparing patients who had 4 or more than 5 coexisting conditions with patients who did not have medical conditions (Table 3).

The preoperative 24-month medical expenditures increased significantly with the incremental number of major preexisting medical conditions among all age groups, and differences among various age groups who had the same number of conditions were relatively minute, although still statistically significant (

). There was a linear increase in average preoperative 24-month expenditures among all age groups: US dollars = n x 600 x constant, where n is the number of coexisting conditions, and the constant is equal to 1.0 to 1.2 (Figure 2B).

After stratifying the preoperative comorbidity-related medical expenditures into low, median, high, and very high levels with equal numbers of patients, there was a consistent positive correlation between the observed postoperative complication rates and the preoperative expenditures for patients’ preexisting illnesses. After adjusting for the preexisting conditions and preoperative medical expenses in each age group, the adjusted ORs for postoperative major complications in the medium, high, and very high medical expenditure subgroups increased from 1.03 (95% CI 0.98-1.08) to 2.01 (95% CI 1.90-2.12) in the group aged 55 to 64 years and increased from 1.17 (95% CI 1.08-1.27) to 1.99 (95% CI 1.81-2.20) in patients aged 

85 years when using the low medical expenditure group as the reference (

). There was overt homogeneity in mortality rates after major complications with adjusted ORs of 1.03 (95% CI 0.85-1.25) to 0.90 (95% CI 0.73-1.11) among different expenditure groups aged 55 to 64 years and adjusted ORs of 0.95 (95% CI 0.80-1.14) to 1.13 (95% CI 0.92-1.37) in the subgroup aged 

85 years compared with the low medical expenditure subgroup, after adjusting for preoperative expenditures in addition to coexisting illnesses (Table 4).

In this large population-based cohort study of elderly surgical patients under a healthcare system with universal coverage, we first verified that the adverse outcomes of inpatient surgery for the elderly were age dependent, related to comorbidities, and associated with preoperative medical expenditures. After adjusting for patients’ covariates, the risk ratios for postoperative major complications, but not for the mortality rates, increased incrementally with numbers of comorbidities and preoperative comorbidity-related medical expenditures among elderly surgical patients. Considering the study scale and patient populations, we validated use of preexisting comorbidities and preoperative medical expenditures to predict the potential surgical adverse outcomes of elderly surgical patients, providing a new way to explore correlations between preconditioned costs (preoperative medical expenditures) and effectiveness (comorbidities or adverse outcomes) in acute care services for the elderly.

Previous studies have primarily focused on either the prognostic impacts of comorbidities or their correlations with medical costs of chronic illnesses in the elderly, without verifying potential correlations between preoperative costs and postoperative outcomes in geriatric surgical patients.

 Theoretically, the prevalence of coexisting medical conditions increases with age, and medical expenses associated with healthcare increase with additional chronic conditions. In this study, we first correlated the number of preexisting medical conditions with major postoperative adverse outcomes and validated their impact in an age-specific pattern. Among various age groups, patients aged 55 to 64 and 65 to 74 years with more than 5 illnesses had higher adjusted ORs (7.23 [95% CI 6.63-7.89] and 6.00 [95% CI 5.63-6.40]) than the patients aged 75 to 84 and 

85 years (4.31 [95% CI 4.04-4.60] and 4.04 [95% CI 3.56-4.49]) for postoperative rates of complications after risk adjustment when using patients without preexisting illnesses as a reference. This result indicates that the complexity of preoperative comorbidities is more influential in younger groups than in older groups in determining the incidence of postoperative adverse outcomes.25 Further analyses (Figure 1B) also confirmed a linear correlation, with complexity (number of illnesses) rather than age differences playing a major role in determining average preoperative expenditures.

 Accordingly, we could estimate potential risk for surgical adverse outcomes based on the number of major comorbidities that elderly surgical patients in different age groups had preoperatively.

Although it might be clinically intuitive to expect patients with more comorbidities to consume more medical resources, analyses among different levels of preoperative medical expenditures exhibited specific correlation patterns. Preoperative medical expenditures showed a positive correlation with the adjusted risk ratio of major postoperative complications, but were irrelevant to the adjusted mortality rate, which was nearly consistent among patients with various levels of cost. Basically, preoperative medical expenditures were for preexisting medical conditions, and healthcare spending was mainly for prescriptions (medications), laboratory examinations, and to a lesser extent, management or procedures related to systemic diseases. Major perioperative complications resulted from major preexisting comorbidities, which also caused subsequent postoperative adverse outcomes. Therefore, coexisting medical conditions, preoperative expenses, and postoperative complication rates were all closely correlated. In contrast, the mortality rate was the final result derived from the combination of patient factors, surgical outcomes, and the quality of postoperative care throughout the period of hospitalization. Influences on the mortality rate were not confined only to patient factors (preoperative comorbidities and associated expenses).

Adjusting for patients' covariates, postoperative complications and mortality among geriatric surgical patients exhibited an age-dependent, illness-related, and preoperative medical expense“associated pattern under universal healthcare coverage.


How Comorbidities and Preoperative Expenditures Correlate With Postoperative Adverse Outcomes

To identify factors associated with physician referrals to end-of-life (EOL) care.

Participants were managed care physicians (n = 545) from the Southern California region of a national nonprofit health maintenance organization who treated a patient in their office within 6 months of the patient’s death from a chronic condition. Measures included self-reports of referrals to EOL services, comfort level discussing EOL with patients, personal/family experience with hospice, and demographic characteristics.

 Participants were most commonly US born (69.1%), married (83.8%), and male (66.0%) with a mean age of 47 years (SD = 8.9 years). About half were Caucasian (51.7%). Logistic regression revealed that family/internal medicine physicians were nearly 9 times more likely to make EOL referrals (95% confidence interval [CI] 3.879-19.434), and physicians comfortable discussing EOL care were nearly 7 times more likely to refer (95% CI 3.465-12.750). Younger age was significantly associated with EOL referrals; with every 1-year decrease in age, physicians were 5% more likely to refer (95% CI 0.911-0.985). Family/internal medicine physicians (95% CI 1.259-2.899) and those comfortable discussing EOL care (95% CI 2.964-9.685) were also more likely to make frequent (4 or more) referrals

 This study highlights factors associated with EOL referrals that may be enhanced at the organizational level through training and educating physicians. Results suggest that organizations should work toward improving physician ease and comfort with EOL conversations. This study serves as an important step toward understanding and reducing physician-level barriers to EOL referrals.

This study is the first to examine physician experience and comfort with end-of-life (EOL) discussions in multivariate models and shows the important contribution these characteristics make.

		Family/internal medicine physicians and comfort conducting EOL conversations were associated with making single and frequent referrals to EOL care.

		Younger physician age was associated with making 1 or more EOL referrals; however, age was not a significant determinant of frequent referrals.

		Results support efforts to reduce physician-level barriers to EOL referrals, support strategies to improve comfort with EOL conversations, and suggest areas organizations might consider to influence physician behavior.

Over the last decade, Americans’ preference to die at home

for end-of-life (EOL) services. In 2009, hospices provided care for 41.6% of US decedents,

 up from 28.6% in 2000.6 Studies indicate that such EOL programs maximize quality of life through palliation of pain and symptoms,

 although initiation and timing of referral may impact the effects of symptom severity caused by disease progression.

 Referrals to EOL services are commonly initiated by physicians, who, for hospice care, must certify that the patient has a terminal condition and estimated life expectancy of 6 months or less. With physicians as gatekeepers,10,11 variation in referral, enrollment, and utilization of EOL services exists, particularly for certain racial groups

 Studies have found that many physicians do not discuss EOL with terminally ill patients. For example, in one study only 53% of stage IV lung cancer patients had had a hospice discussion with their provider.

 Physicians have identified discomfort with death and dying, including the hospice concept, as a barrier to initiating EOL discussions and hospice referrals.

Research on EOL referrals has tended to focus on patient-level characteristics rather than physician factors. Among the few studies of physician behavior, results are mixed; some indicate that age,

 impact EOL referral rates, whereas others do not.

These results highlight the need to better understand how physicians’ personal characteristics (eg, sex, age, race, area of practice) influence EOL referrals to hospice and palliative care. Additionally, studies have found that hospice enrollment is affected by familiarity with hospice because of having had a close friend or relative under hospice care

; however, this factor has not been explored among physicians. Although previous studies have analyzed individual physician barriers to EOL referrals, physicians’ personal experience and comfort have been largely unexamined. To address this gap, we explored the impact of multiple physician-level variables, adjusting for fixed demographic characteristics, on referrals to hospice among a sample of Southern California-based physicians from a large health maintenance organization (HMO).

We conducted a cross-sectional study of Southern California-based physicians from a large HMO. Potential participants included all HMO physicians practicing in the specified geographical region who had seen a patient in their office within 6 months of the patient’s death from a chronic condition. Individuals were contacted via e-mail by the physician co-principal investigator and were given a link to an anonymous web-based survey. All were first contacted in September 2008; nonresponders were sent reminder e-mails in each of the next 2 months. This study was approved by the institutional review boards of the participating institutions.

We developed a 43-item survey that included physicians’ self-reported number of patient referrals to EOL services (home-based palliative care and hospice), knowledge of EOL services, attitudes toward EOL services, perceived barriers to referring patients to EOL services, personal and family experience with utilizing EOL services, and demographic characteristics. Prior to administration, the survey had been reviewed for face validity, comprehension, and completeness by the research team and then pretested by 2 physicians.

Home-based palliative care is provided by this HMO for seriously ill patients with cancer, congestive heart failure, and chronic obstructive pulmonary disease who are expected to have about 1 year to live. The service is similar to hospice with the exception that patients do not have to forgo curative measures, and life expectancy may exceed 6 months (see Brumley et al

). Physicians’ EOL referral behaviors (dependent variable) were assessed based on self-reports in response to the following survey item: Estimate the number of patients you refer to hospice or home-based palliative care on an annual basis.

Rate of referral was recoded into 2 separate dichotomous variables to analyze differences between referrers and nonreferrers to EOL care and between those who made frequent (4 or more) referrals as opposed to no referrals. Frequent referrals were defined as 4 or more annual referrals because the original variable was categorical (none, 1-3 referrals, 4-6 referrals, 7-9 referrals, or 10 or more referrals).

Physicians were asked to record their sex, race, marital status, age, nativity, and length of time practicing medicine (postresidency). Additionally, physicians were asked to select their area of medical practice from a list of common areas of practice or to indicate their specialty in an open-ended field.

Physicians’ comfort level discussing EOL services with terminally ill patients was measured on a 5-point Likert scale in response to the statement: 

I feel comfortable discussing hospice and palliative care options with terminally ill patients during outpatient visits

. The response set ranged from “strongly disagree” to “strongly agree,” coded with a higher score reflecting stronger agreement.

Physicians were asked to indicate whether they had a family member who had received hospice care.

Descriptive statistics were used to describe key characteristics of the sample, and c2 tests were conducted to analyze differences between physicians’ EOL referral behaviors and demographic characteristics. We performed t tests to examine variations in mean age between referrers and nonreferrers to EOL care and between those who made frequent referrals versus those who made no referrals. Logistic regressions were conducted to identify predictors of any EOL referral and frequent referrals. In each condition, 3 models were entered: model 1 (a and b) fit demographic variables, model 2 (a and b) included all variables in model 1 and physicians’ comfort with conducting EOL discussions with patients, and model 3 (a and b) included all variables in models 1 and 2 along with physicians’ personal/family experience with hospice. Based on comparisons in the c2 change statistic, the most parsimonious model was selected. SPSS statistical software for Windows version 18.0 (SPSS Inc, Chicago, Illinois; released 2009) was used in all analyses.

Between September and November 2009, 2084 physicians were contacted; of these, 554 completed the survey (response rate = 26.65%); 1525 were unresponsive to the emails; and 5 opted out of the study. Most respondents were US born (69.1%), married (83.8%), and male (66.0%), with a mean age of 47 years (standard deviation = 8.9 years). About half (51.7%) were Caucasian, 31.2% Asian, 8.7% Latino, 3.3% multiethnic, 3.1% African American, and 1.1% were other races. Of those born outside the United States, more than half were from Asia (57.1%) followed by North America (Canada and Mexico) (11.4%), the Middle East (9.3%), Africa (8.6%), Europe (7.9%), and South America (5.7%). Area of practice was most commonly a specialty (46.2%) or family/internal medicine (45.6%). Specialty areas included surgery (24.3%), oncology (11.7%), and psychiatry (11.3%) among 26 reported specialties. Fewer than half (43%) reported that a family member received hospice care, and most (83.2%) reported making at least 1 EOL referral in the last year and feeling comfortable (80.4%) conducting EOL discussions with terminally ill patients (

). Half (50.5%) reported making 4 or more referrals in the last year.

Area of practice (χ2 = 45.748, P <.001) and comfort conducting EOL conversations (χ2 = 45.965, 

 <.001) varied significantly between physicians who made an EOL referral and those who did not. The number of years practicing medicine was also significantly different between EOL referring and nonreferring physicians (χ2 = 13.974, 

 = .007). Differences in physician mean age revealed a significant inverse relationship, such that younger physicians were more likely to report referring (mean age of referring physicians = 46.39 years, SD = 8.75 years; mean age of nonreferring physicians = 49.79 years, SD = 8.76 years; 

 <.001; Table 1). There were no differences in sex, ethnicity, marital status, nativity, or personal experience in referral to EOL care.

Logistic regression showed that family/internal medicine physicians and comfort with conducting EOL conversations with patients were significantly associated with making a referral. Family/internal medicine physicians were nearly 9 times more likely to make EOL referrals (95% confidence interval [CI] 3.879-19.434) compared with specialists and emergency medicine physicians. Physicians reporting comfort with discussing EOL care were nearly 7 times more likely to refer patients compared with physicians who were uncomfortable conducting these discussions (95% CI 3.465-12.750) (

 model 1a). Age was a significant negative predictor of EOL referrals; with every 1-year decrease in age, physicians were .05 times more likely to make an EOL referral (95% CI 0.911-0.985). Other demographic variables and having had a family member receive hospice care were not significant predictors.

The second regression examined factors associated with frequent EOL referrals (also adjusted for demographic and practice variables) and identified family/internal medicine physicians and comfort in conducting EOL as significantly associated with frequent referrals to hospice. Family/internal medicine physicians were nearly 2 times more likely to make frequent EOL referrals (95% CI 1.259-2.899) compared with specialists and emergency care physicians, and those comfortable discussing EOL care were 5 times more likely to frequently refer compared with participants who reported a lack of comfort conducting these discussions (95% CI 2.964-9.685) (

, model 2b). Demographic variables including age and having a family member who received hospice care were not significant predictors of frequent EOL referrals.

This study identified physician characteristics associated with referrals to hospice or palliative care. Respondents had a high rate of referral (83%) and expressed comfort conducting EOL conversations (80%). Comfort was positively associated with making single and frequent referrals. Previous studies on physician factors and EOL care comparing populationbased hospitals, HMOs, and Veterans Health Administration (VHA) sites found that physicians from HMOs and VHA managed care models reported conducting EOL conversations earlier in the disease process than physicians in populationbased settings,

 and that HMO patients enrolled in hospice at higher rates than patients in fee-for-service plans (17% vs 11%, respectively).

 Thus, the institutional environment may positively influence referral practices.

Practicing in the area of family/internal medicine was associated with making single and frequent referrals after adjusting for other factors. This finding reflects practice variations between family/internal and specialty physicians, which was found in previous studies,

 or a closer relationship between family/internal physicians and their patients, fostering greater comfort with sensitive discussions. Physician age was negatively associated with any EOL referral (but not frequent referrals), supporting previous studies, which found younger age

 or fewer years in practice (less than 10 years)

 to be associated with EOL referral. This may reflect an age-related cohort effect from the mandated inclusion of EOL care training and education in accredited US medical schools in the past 13 years.

This study used multivariate models to identify physician specialty and comfort with end-of-life discussions, both of which are associated with end-of-life referrals.

Physician Factors That Influence Patient Referrals to End-of-Life Care

While active surveillance for adverse drug events using electronic medical records (EMRs) and claims data is expanding, these data do not fully capture patient experiences with medication-related symptoms. Our objective was to describe adherence and outcomes associated with an automated pharmacovigilance call.

Prospective cohort of patients receiving a prescription for a target medication at a participating primary care clinic were eligible for an automated phone pharmacovigilance call. Outcomes, compared for participants who completed (n = 1184) versus did not complete the call (n = 407), included EMR documentation during the 6 months following the call that the patient had: stopped the medication, used acute care services or died, or received a specialty or primary care visit.

Compared with those who agreed to participate but did not complete the pharmacovigilance call, subjects who completed the call had greater rates of EMR-documented medication cessation (3.9% vs 1.0%, adjusted 

 value = .007), and use of primary or specialty care (32.8% vs 18.7%, 

 value <.0001), but similar use of acute care services or death (12.2% vs 9.8%,

 = .38). Of participants, 50.2% reported 

1 symptom; of these, 22.0% thought the symptom was medication-related. In contrast to the low rates of EMR-documented medication cessation, 21.2% of participants said that they were no longer taking the medication; 69% said that their doctor did not know that they had stopped.

Automated phone pharmacovigilance provides important information about adherence, and was associated with greater EMR-documented medication cessation and planned service use.

Routine automated phone pharmacovigilance can provide important information about medication adherence and was associated with documentation of medication cessation, and use of outpatient care. One-half of patients experienced >1 symptom and almost one-fourth attributed the symptom to the target medication.The safety of prescription drugs represents a major public health concern. Estimates suggest that approximately 0.5% of all emergency department (ED) visits and primary care visits result from an adverse drug event (ADE).

While the US Food and Drug Administration (FDA) maintains a voluntary ADE reporting system for post-marketing surveillance, only 1% of ADEs are reported.

 Among the reasons for this underreporting are the difficulty that patients and providers may have linking a particular symptom or condition to a specific drug, particularly for individuals who have chronic illness and are taking several medications, a common clinical scenario in primary care. While the capacity for active surveillance using electronic medical records (EMRs) and claims data is expanding,

 these data will not fully capture the patient experience, as clinicians often fail to completely document a patient’s symptoms.

Interactive voice response system (IVRS) technology has the potential to systematically and efficiently contact large numbers of patients directly to assess adherence as well as symptoms. We previously described an IVRS system linked to our EMR, used to contact patients recently prescribed certain prescription medications to assess their symptoms in primary care practice.

 The goal of this study was to see whether participation in this type of systematic monitoring system was associated with subsequent changes in medication management or healthcare utilization. Our primary hypothesis was that patients who participated and completed our e-pharmacovigilance call would be more likely to have their mediations stopped than those who did not complete the call, as these patients would be more likely to have symptoms identified attributable to their medications. The realization that their symptom may be related to a medication would result in more planned outpatient visits to primary care physicians or specialists but fewer unplanned acute care visits (eg, ED or hospitalization) or death because of earlier detection of problems. Our secondary hypothesis was that these effects would be mediated by the reporting of medication symptoms but, since symptom reporting was only available for participants, we addressed this hypothesis by comparing the same outcomes as above between participants who did and did not report medication symptoms. This analysis provides some of the first data on the implications of active pharmacovigilance on healthcare utilization and outcomes.

We conducted the study in 11 primary care clinics affiliated with the Brigham and Women’s Primary Care Practice- Based Research Network. Adult, English-speaking patients were eligible if they had recently been started on 1 of the target medications listed in the Protocol section by a participating primary care physician (PCP). Patients with a visit during 2008 to 2010 were eligible. The study was reviewed and approved by the Partners Healthcare Institutional Review Board.

Our protocol has been described previously.

 In brief, potentially eligible patients prescribed selected medications were identified from our EMR. Target medications were selected based on the date of FDA approval, as drugs are more likely to have unrecognized adverse events noted within 7 years of introduction to the market,

 and/or particular concerns about side effects. Our goal was to move beyond drugs that have been commonly associated with ADEs resulting in ED visits, like warfarin and hypoglycemic agents,

 to identify drugs that may cause more chronic disability and symptoms that interfere with daily life which may otherwise not be recognized. Drugs selected were: zolpidem, eszopiclone, sildenafil, vardenafil, tadalafil, losartan, valsartan, irbesartan, aliskerien, gabapentin, pregabalin, montelukast, fluticasone propionate-salmeterol, varenicline, rosuvastatin, ezetimibesimvastatin, rosiglitazone, pioglitazone, exenatide, sitagliptin, risperidone, olanzapine, quetiapine, modafinil, aripiprazole, duloxetine, raloxifene, trospium, infliximab, ibandronate, and etanercept.

 Patients who did not opt-out after receiving an informational letter (91.1%) were called approximately 4 weeks following their primary care visit using IVRS (CallAssure, Vocantas, Inc). Overall, 22.5% of patients with a working phone number (30.6% of individuals in a household where a live individual answered the phone) completed the IVRS tool.

A standard script with branching logic first asked each patient whether they were taking the target drug. If the patient stated that they were not taking the drug, they were asked whether their doctor knew that they had stopped taking the medication, and specific reasons for stopping (including whether they experienced a side effect or symptom). If they were no longer taking the target medication and it was not stopped because of a side effect or symptom the survey was concluded. If they reported that they had stopped the medication for a side effect or symptom, they were then asked a standard checklist of symptoms that are potentially associated with ADEs among ambulatory patients.

 Patients who stated that they were still taking the medication were asked whether they experienced any of the symptoms on the checklist since their most recent PCP visit. If an individual reported a symptom, they were asked if they thought that the symptom was related to the target medication (coded as yes, no, don’t know, or missing response). Information reported during the call was automatically documented for all patients in the EMR using a templated note. In addition, the system sent a single e-mail to the PCP if the patient: (1) reported that they had stopped taking a medication intended for chronic use and said they had not discussed this with their provider, (2) reported a pre-defined list of symptoms, specific to each drug, that were of clinical concern, or (3) requested that we send an e-mail to their provider.

The goal of this study was to evaluate the association between completion of the e-pharmacovigilance call and subsequent healthcare utilization and mortality.

The primary outcomes assessed included whether an individual: (1) stopped the target medication, as documented in the EMR by a healthcare provider; (2) visited a PCP or specialist physician; or (3) experienced a combined outcome of ED visit, hospitalization, or death. These 3 outcomes were assessed within 6 months of the date that the patient’s phone number was uploaded to the IVRS server, the “eligibility date.” As secondary outcomes, we compared self-reported with EMR-documented medication cessation, and the primary study outcomes listed above for those who participated and reported versus did not report a symptom.

While completion of the e-pharmacovigilance call was our primary predictor, we were concerned about potential confounding effects due to clinical and demographic differences between non-participants, and participants who completed versus did not complete the call. Therefore, we collected and analytically adjusted for patient characteristics that were potentially associated with the outcomes, including age (continuous), gender, race/ethnicity (white, black, Hispanic, other), count of comorbid conditions including cancer, dementia, diabetes, heart problems, hypertension, elevated lipids, osteoporosis, vascular disease, and kidney, liver, lung, and psychiatric problems (0, 1, 2, 3, 4 or more), drug class (insomnia, erectile dysfunction, angiotensin receptor blockers, seizure, asthma, smoking cessation, lipid lowering, diabetes, psychiatric, other) and median household income of the patient’s zip code and number of ED visits and hospitalizations (0, 1-5, 6 or more) and number of primary care and specialist visits (continuous) in the year prior to enrollment.

We used bivariate analysis to examine the associations of completing the pharmacovigilance call with patient characteristics; χ2 tests were used to determine statistical significance. We also compared those individuals who agreed to participate with those who were potentially eligible but did not participate. To compare healthcare utilization by completion status, we used multilevel logistic regression models, clustered on primary care physician using SAS version 9.2 (SAS Institute, Cary, North Carolina). Multilevel models were then adjusted for patient characteristics including age, gender, race/ethnicity, count of comorbid conditions, drug class, median household income, prior ED visits and hospitalizations, and primary care and specialist visits, and also adjusted for primary care clinic.

The median age of individuals who completed the call as 59 years, compared with 60 years for those who agreed to participate but did not complete the call and 56 years for those who did not participate (

). Women were more likely to complete the call. Compared with those who did not participate, individuals who participated in the e-pharmacovigilance system were more likely to be white, live in a zip code with a higher household income, and have more comorbid conditions than those who did not participate. Of the individuals who participated in the e-pharmacovigilance call, 799  (50.2%) reported at least 1 symptom. Of the 261 participants whose call generated an e-mail to their provider (16.4% of all participants), 10.0% indicated that the patient was no longer taking the study medication, 28.0% because the patient requested that their provider be notified, and 62% because they reported a pre-specified symptom.

Outcomes Associated With Participation in e-Pharmacovigilance

While only 3.9% of patients who completed the e-pharmacovigilance call and 1.0% of patients who did not complete the call had discontinuation of the target medication recorded in the EMR within 6 months of their eligibility date, this difference was statistically significant (

In contrast to the low rates of medication cessation documented in the EMR, 21.2% of participants said that they were no longer taking the medication at the time of the pharmacovigilance phone call. Self-reported medication cessation was higher than cessation documented in the EMR for all drugs, including those used for diabetes (12.8%) and lipids (8.0%), but the self-reported rate was highest for varenicline (39.5%). Of those who said that they were no longer taking the medication, 10.1% said it was because they had had at least 1 symptom, 7.1 % because of cost, 11.8% because they thought the medication did not work, and 27.2% because they thought that the medication was no longer necessary. Sixty-nine percent said that their doctor did not know that they had stopped the medication, consistent with the lower rates of documentation in the EMR.

Rates of the combined outcome of ED visit, hospitalization, or death were not significantly different between participants who completed versus did not complete the call (12.2% vs 9.8%; 

 = .38). Rates of primary or specialty care visits were higher for participants who completed versus did not complete the call (32.8% vs 18.7%, 

Among the patients who participated in the e-pharmacovigilance call, those who experienced a symptom were much more likely than those who did not report a symptom to have documentation in the EMR that their medication stopped (4.5% vs 1.8%, adjusted 

). Patients who reported a symptom were more likely to have a primary or specialty care visit (34.8% vs 23.5%, 

 = .0005), but there was no difference in the combined outcome of ED visit, hospitalization, or death (13.9% vs 9.2%, 

Patient’s Assessment of Whether a Symptom Was Associated With the Target Medication

Among the sub-group of 799 participants who noted at least 1 symptom, 176 (22.0%) attributed their symptom to the medication, 406 (50.8%) thought that the medication was not related to the symptom, and the remaining 217 (27.2%) did not know or did not report whether there was an association. Patients who thought that their symptom was related to the medication were more likely to report that they had stopped the medication (13.1%) compared with those who did not think that the symptom was related to the medication (1.5%) or who did not respond (3.0%, 

This study provides important data on the potential implications of a new approach to active pharmacovigilance on healthcare utilization and outcomes. Supporting our hypothesis, completion of this e-pharmacovigilance call was associated with a difference in EMR-documented medication cessation and greater use of planned visits, but we did not see a decline in unanticipated healthcare. Importantly we also found that rates of documentation in the EMR were substantially lower than the rates of discontinuation noted by the patients, suggesting that this may be an important tool for improving physician-patient communication about medication adherence. Of patients who reported a symptom on the e-pharmacovigilance call, one-fourth thought that the symptom was related to the medication and one-third did not know whether there was an association. This finding highlights both how difficult it can be for patients to decide to attribute a symptom to a drug, a necessary for step for ADE reporting, but also suggests that many patients think that they are experiencing a medication-related symptom. Ascertaining information directly from patients is crucial to prevent potential medication-related morbidity and is also necessary to improve adherence when a symptom is unlikely to be related to a particular medication. The large discrepancy between self-reported medication cessation and documentation in the EMR also underscores the importance of adding the patient perspective to safety studies that are based solely on claims or chart review.

Few studies have integrated the patient perspective in the surveillance of ADEs.

 A phone survey of outpatients from 4 primary care clinics in the United States who received at least 1 prescription drug during a 1-month period found that 25% reported an ADE, but did not look at subsequent utilization.

This study highlights the potential value of innovative ways of collecting information about adverse drug events directly from patients.

November 2012

Care Coordination Agreements: Barriers, Facilitators, and Lessons Learned

How Comorbidities and Preoperative Expenditures Correlate With Postoperative Adverse Outcomes

Physician Factors That Influence Patient Referrals to End-of-Life Care

Active Pharmacovigilance and Healthcare Utilization

A Systematic Review of Reference Pricing: Implications for US Prescription Drug Spending

Letter to the Editors: Response to Payer Coverage for Patients Enrolled Onto Clinical Trials

Patient-Centered Medical Home Cost Reductions Limited to Complex Patients

Automated Telephone Calls to Enhance Colorectal Cancer Screening: Economic Analysis

Utilization of HER2 Genetic Testing in a Multi-Institutional Observational Study

Validating the Adapted Diabetes Complications Severity Index in Claims Data

Hospital Admissions and MS: Temporal Trends and Patient Characteristics

Value-Based Purchasing