Continuity of Care and Health Care Utilization in Older Adults With Dementia in Fee-for-Service Medicare

Importance  Poor continuity of care may contribute to high health care spending and adverse patient outcomes in dementia.

Objective  To examine the association between medical clinician continuity and health care utilization, testing, and spending in older adults with dementia.

Design, Setting, and Participants  This was a study of an observational retrospective cohort from the 2012 national sample in fee-for-service Medicare, conducted from July to December 2015, using inverse probability weighted analysis. A total of 1 416 369 continuously enrolled, community-dwelling, fee-for-service Medicare beneficiaries 65 years or older with a claims-based dementia diagnosis and at least 4 ambulatory visits in 2012 were included.

Exposures  Continuity of care score measured on patient visits across physicians over 12 months. A higher continuity score is assigned to visit patterns in which a larger share of the patient’s total visits are with fewer clinicians. Score range from 0 to 1 was examined in low-, medium-, and high-continuity tertiles.

Main Outcomes and Measures  Outcomes include all-cause hospitalization, ambulatory care sensitive condition hospitalization, emergency department visit, imaging, and laboratory testing (computed tomographic [CT] scan of the head, chest radiography, urinalysis, and urine culture), and health care spending (overall, hospital and skilled nursing facility, and physician).

Results  Beneficiaries with dementia who had lower levels of continuity of care were younger, had a higher income, and had more comorbid medical conditions. Almost 50% of patients had at least 1 hospitalization and emergency department visit during the year. Utilization was lower with increasing level of continuity. Specifically comparing the highest- vs lowest-continuity groups, annual rates per beneficiary of hospitalization (0.83 vs 0.88), emergency department visits (0.84 vs 0.99), CT scan of the head (0.71 vs 0.83), urinalysis (0.72 vs 1.09), and health care spending (total spending, $22 004 vs $24 371) were higher with lower continuity even after accounting for sociodemographic factors and comorbidity burden (P < .001 for all comparisons). The rate of ambulatory care sensitive condition hospitalization was similar across continuity groups.

Conclusions and Relevance  Among older fee-for-service Medicare beneficiaries with a dementia diagnosis, lower continuity of care is associated with higher rates of hospitalization, emergency department visits, testing, and health care spending. Further research into these relationships, including potentially relevant clinical, clinician, and systems factors, can inform whether improving continuity of care in this population may benefit patients and the wider health system.

The growing population of older adults with dementia poses a unique challenge to the US health care system.¹ The cost of caring for this population is on par or higher than the financial burden of heart disease and cancer.^2,3 Individuals with dementia have high rates of hospitalization, including potentially preventable hospitalization,^4-7 and are frequently evaluated in outpatient or emergency department (ED) settings.^6,7 Hospitalization and emergent evaluations increase medical costs⁶ and, more importantly, may subject persons with dementia, who frequently have multiple chronic conditions, to distressing and potentially harmful interventions given risk of adverse outcomes in this population.^8-12 Understanding key factors affecting medical care in dementia is required to promote more efficient, effective, and patient-centered health care delivery.

Most persons with dementia receive treatment for this complex neuropsychiatric illness and medical comorbidities in the ambulatory setting.^13,14 Ambulatory care in the United States is often fragmented. The average Medicare beneficiary sees 7 physicians in 4 different practices annually, and communication and coordination between physicians is generally suboptimal.^15,16 Quality of care in dementia is also threatened by other factors, including clinician time constraints, clinician comfort, impaired patient communication and insight, and overwhelmed caregivers.^16-18 Continuity of care (COC) at the level of the clinician may be particularly important for building clinician-patient-family relationships, addressing goals and expectations of care over time, understanding patients’ cognition and stage of dementia, and recognizing and appropriately managing acute and chronic conditions in this population.

To our knowledge, COC in dementia has not previously been studied despite its potentially crucial role in health care for persons with this complex, expensive illness. In other populations, higher COC is associated with decreased hospitalization, medical procedure overuse, and cost.^19-23 Greater COC may also improve patient-clinician trust,²⁴ quality of communication,²⁵ and patient satisfaction,^23,26,27 factors that may alleviate barriers to high-quality care in dementia.^16-18 Recent health care reforms promote COC through accountable care organizations and patient-centered medical homes, models of care that emphasize care coordination.^22,28,29 The anticipated effects of such reforms on improving care in dementia are not fully understood. The objective of our study was to examine and understand the association of COC with health care utilization and spending in the vulnerable population of community-dwelling older adults with dementia.

Box Section Ref ID

We used the complete national sample of fee-for-service patients insured by Medicare in 2012. We first identified Medicare beneficiaries who were 65 years or older on January 1, 2012, and had 12 months of continuous Parts A and B coverage from the 2012 Medicare Beneficiary Summary file. We then searched their 2012 inpatient (Medicare Provider Analysis and Review file [MedPAR]) and outpatient (Physician/Supplier Part B file and Outpatient file for rural or federally qualified health centers) claim records to identify diagnoses of dementia (see eTable 1 in the Supplement for diagnostic codes). Patients needed 1 outpatient or inpatient dementia claim in 2012 to be considered to have the condition.^30,31 We restricted study patients to those who had a claims diagnosis of dementia (n = 2 657 648 patients), lived in the community (631 544 patients with >100 nursing home days according to Minimum Data set were excluded), survived the entire year (207 028 patients who died were excluded), and had at least 4 outpatient visits in 2012 to allow for calculation of the continuity measure (402 707 patients with <4 visits were excluded ). All remaining patients were included in the study (1 416 369 patients). The data were analyzed from July to December 2015. The Dartmouth College Institutional Review Board approved this study.

Continuity of care reflects the degree to which patient visits are concentrated among clinicians. The COC score measures the concentration of a patient’s visit pattern, assigning a higher score to visit patterns in which a larger share of the patient’s total visits are with fewer clinicians (see eTable 2 in the Supplement for an example of score derivations).^32,33 The COC score was calculated using a patient’s ambulatory evaluation and management visits with physicians or nurse practitioners in 2012. The formula

(∑ n_i² – N)/N (N – 1),where n_i = number of visits that the patient has with the ith physician and N = total visits, has been used previously.^22,32,34 Each National Provider Identifier in outpatient claims was considered a unique clinician. The COC index is less meaningful with few visits because it is easy to attain a minimum score of 0 or maximum score of 1, so we restricted analyses to patients with at least 4 visits,^34-36 which represented approximately 85% of all older beneficiaries with a dementia diagnosis. The COC scores were converted from a continuous value to low, medium, and high tertiles based on the distribution of scores for analyses and ease of interpretation because the actual score lacks inherent clinical meaning.^19,35

We measured 4 categories of health utilization and spending in 2012: hospitalizations, ED visits, imaging and laboratory testing, and health care spending.

1. Hospitalizations: We identified acute, nonobservation, short-stay hospitalizations from acute care or critical access hospitals from inpatient claims. We further measured hospitalizations for ambulatory care sensitive conditions (ACSC) defined by the Agency for Healthcare Research and Quality as prevention quality indicators. ACSC hospitalizations represent conditions for which hospitalization could be avoided if the patient receives timely and adequate outpatient care, such as diabetes complications, chronic obstructive pulmonary disease (COPD) or asthma, hypertension, congestive heart failure (CHF), dehydration, pneumonia, and urinary tract infection. We examined composite ACSC hospitalizations as well as acute and chronic ACSC hospitalizations subcategories.³⁷

2. Emergency department visits: We identified ED visits that did not result in inpatient hospitalization from outpatient claims; ED visits leading to hospitalization were captured in the hospitalization outcomes.

3. Imaging and laboratory testing: While persons with dementia and other chronic conditions undergo the same routine laboratory tests and imaging as individuals without dementia, certain tests may be overused in this population to evaluate changes in mental status or behavioral symptoms. Specifically, computed tomographic (CT) scan of the head, chest radiography, urinalysis, and urine culture may be ordered to evaluate for an infection or acute neurologic event in the absence of a suggestive history or localizing signs and/or symptoms.^38,39 We identified these specific imaging and laboratory tests from inpatient and outpatient claims (see eTable 3 in the Supplement for Current Procedural Terminology codes).

4. Health care spending: We examined Medicare spending in 3 categories. Hospital and skilled nursing facility spending from the MedPAR file reflects inpatient spending, while physician spending was derived from Medicare Part B spending. Total spending includes MedPAR, Part B claims, home health, hospice, durable medical equipment, and other facility spending. We standardized spending to adjust for differences in Medicare reimbursement due to cost-of-living, disproportionate share, graduate medical education, and hospital payments.⁴⁰

From the denominator file, we obtained beneficiaries’ age, sex, race/ethnicity, Medicare-Medicaid dual eligibility, and ZIP code of residence. We linked 9-digit ZIP codes to the 2010 US Census Tract to obtain median household income. ZIP code was also linked to hospital referral region to consider regional market-related characteristics.⁴¹ We accounted for illness burden using the Hierarchical Condition Categories (HCC) score, a Medicare risk-adjustment system that gives more weight to comorbidities and severe manifestations of comorbidities that have greater effect on health care utilization.⁴² We also examined the total number of ambulatory visits, primary care (internal/family medicine, geriatrics, nurse practitioners) and specialist (all other specialties) visits, unique outpatient clinicians seen, and predominant ambulatory clinician in 2012 to better understand COC score components; these variables were not included in statistical models given their representation within the COC score.

We compared characteristics of the study population by COC tertile to determine if there were significant differences among the 3 groups. To account for observed differences between each of the COC tertiles, we applied propensity weighting methods. Inverse probability weighting was used to balance the differences in patient characteristics across continuity levels. We used a multinomial logistic regression model to estimate the probability of each beneficiary belonging to their actual COC tertile based on their observed characteristics. Age, sex, race, HCC score, dual eligible status, and median household income were included as covariates in the model based on their potential to be predictive of COC level; HCC score and dual eligible status were the strongest predictors of COC level. In sensitivity analysis, hospital referral region was removed from the final model given it did not add significantly to balancing covariates between tertiles. The inverse propensity score, or inverse of the probability of each patient being in their actual COC level, was then calculated. These weights were applied to the characteristics to obtain weighted means and counts to assess the balance of covariates after risk adjustment; no more than 10% absolute difference between the 3 COC groups was required. For each outcome, the inverse probability weight was then applied to generate a weighted mean or rate. Based on the weighted results, differences between COC tertiles were evaluated using analysis of variance. Analyses were conducted using SAS statistical software (version 9.3; SAS Institute Inc).

Characteristics of the 1 416 369 community-dwelling Medicare beneficiaries who had a claims diagnosis of dementia, survived the year, and had 4 or more ambulatory visits in 2012 are shown in Table 1. Their mean age was 81 years, with 63.3% female and 82.9% white beneficiaries. The sample had a mean of 13.6 outpatient visits with 4.8 unique clinicians in the year. With increasing levels of continuity, age, proportions of women and nonwhites were higher, and median household income was lower. Individuals with low continuity had more chronic conditions and higher HCC scores; they had higher proportions of coronary artery disease, CHF, and COPD compared with those with high continuity. The low-continuity group had a mean of 15.6 visits to 7.1 unique clinicians, compared with 14.8 visits to 4.8 clinicians and 10.5 visits to 2.5 clinicians in the medium- and high-continuity groups, respectively.

The inverse probability weighted sample included 1 416 344 beneficiaries (25 beneficiaries were excluded owing to missing median household income). After inverse probability weighting, key characteristics became more similar between individuals with different levels of continuity (Table 2). Mean age, HCC score, income, sex, race, and dual eligible proportions were comparable among groups. Mean chronic conditions and coronary artery disease and CHF proportions also became more similar. Mean outpatient visits remained different, with almost 15 visits annually in the low-continuity group vs 11 visits in the high-continuity group. The low-continuity group saw more unique physicians.

Table 3 shows unweighted proportions of health services utilization by continuity level in 2012. Overall, 47% of the sample had at least 1 hospitalization, and 47% had at least 1 ED visit in the year; 47% also had a CT scan of the head performed; and 13% of the sample experienced at least 1 ACSC hospitalization. Crude rates of hospitalization, ED visits, testing, and mean health care spending were highest in the low-continuity group, followed by the medium- and high-continuity groups, respectively. The mean amount of health care spending per beneficiary in 2012 was $22 740. Approximately 55% of total spending per beneficiary was due to hospital or skilled nursing facility spending.

As shown in the Figure, A, and eTable 4 in the Supplement, even after inverse probability weighting, the use of most health services was higher with lower continuity. The annual rate of hospitalization per beneficiary was 5.8% higher and ED visits were 15.4% higher in the lowest continuity compared with highest continuity group. While a CT scan of the head and urinalysis were performed in the low-continuity group at higher rates, chest radiographs were performed at similarly high rates across groups, with over 2 chest radiographs per beneficiary at all continuity levels. ACSC hospitalizations were relatively similar across continuity groups, including when acute and chronic ACSC hospitalizations were examined separately. Significant differences in health care spending remained (Figure, B). Mean yearly spending was $2367 higher in the lowest- vs highest-continuity groups, with higher inpatient and outpatient costs. Between-group differences were statistically significant for all outcomes (P < .001 for all comparisons).

We explored the possibility of interaction effects between predominant clinician and COC tertile on each outcome (eTable 5 in the Supplement) given potentially different effects of continuity in primary vs specialty predominant care. Using a 2-way analysis of variance, the interaction was significant (P < .05) for all outcomes. Thus, effects of continuity differ between those whose predominant clinician was primary care vs a specialist.

Low continuity of ambulatory care among community-dwelling older adults with a dementia diagnosis is associated with higher rates of hospitalization, ED visits, radiologic and laboratory testing, and greater health care spending. Lower continuity is not associated with more ACSC hospitalizations in this population, however. The overall volume of health services utilization and testing is striking, with almost half of the cohort experiencing hospitalization, an ED visit, and a CT scan of the head in the course of the year. Rates of testing translate to approximately 2 chest radiographs and 1 urinalysis per beneficiary per year. When considering differences in health care spending per beneficiary, individuals with the most fragmented care are associated with an additional $567 million to $1.1 billion in health care spending compared with those with medium or high continuity. Within care continuity, the balance of primary care and specialists may also be important, given this population is particularly sick and complex.

These findings resemble patterns seen in studies of COC in other populations; findings regarding potentially avoidable hospitalizations, however, have been mixed. Odds of hospitalization among older adults in Taiwan were 38% and 68% lower with medium and high continuity, respectively, compared with low continuity. The same study found a similar relationship for avoidable hospitalizations, which contradicts our finding of no clinically significant difference in ACSC hospitalizations.¹⁹ A study in younger Medicaid beneficiaries found that higher clinician continuity was associated with lower likelihood of hospitalization, with an 8% to 11% absolute difference between the highest- and lowest-continuity groups in proportion of patients hospitalized; there was no significant association with acute ACSC hospitalization.²⁰ Other studies in older adults, however, found that higher continuity was associated with lower preventable hospitalization rates.^34,43 Higher continuity was also associated with a decrease in potentially overused procedures in Medicare beneficiaries.²² Imaging and laboratory tests examined in our study may similarly represent overused procedures in persons with dementia. Our results are also similar to findings in older veterans and other chronic conditions.^21,43 Medicare beneficiaries with CHF, COPD, or diabetes had lower odds of hospitalization, ED visits, complications, and lower cost with higher levels of continuity.²¹ Compared with those beneficiaries, the mean HCC score in our cohort (1.64) was higher than the median HCC scores for beneficiaries with CHF (0.7), COPD (0.6), and diabetes (0.5). Annual utilization was also higher in dementia compared with proportions with hospitalization or ED visits in CHF (10.5% and 44.6%) and COPD (6.8% and 35.9%).²¹

While our findings reflect the importance of COC found in other populations and conditions, the importance for older adults with dementia may extend beyond health care utilization. Quality of care in dementia has been noted to be inconsistent and often reactive.⁴⁴ Persons with dementia are at risk of experiencing adverse events during the course of medical care.^{8,10,11,45,46} This population is also at risk of unnecessary testing, which can lead to patient burden through invasive tests and overtreatment.^47,48 Administration of sedative-hypnotic medications to complete a CT scan or catheterization for a urine sample are examples of clinical circumstances that may be particularly challenging in persons with dementia. Our results suggest the possibility that addressing care fragmentation could decrease hospitalizations, ED visits, unnecessary testing, and overtreatment in this at-risk population. Potential mechanisms include clinician familiarity with patients’ cognitive and functional abilities, coexisting conditions, support systems, and goals of care. Decreased fragmentation may facilitate anticipatory guidance and communication/coordination between clinicians. Beneficiaries with greater continuity also had greater primary care involvement; for complex patients, balance between primary care and select specialists, with clinician continuity, may prove most beneficial. Interestingly, rates of ACSC hospitalizations were similar across continuity levels. It may be that these conditions present differently in patients with dementia, as symptoms specific to the ACSC condition may be overshadowed by delirium, and early signs and/or symptoms are missed even with high continuity and a longstanding physician relationship. Conflicting results may also point to shortcomings in the ACSC construct, originally designed as a metric of area-level access to quality care.⁴⁹

This study has several limitations. First, we used claims-based dementia diagnoses, which may not represent the entire population of patients with dementia.^30,31 Individuals with Medicare managed care, for example, were excluded; people with mild disease may be missed; and individuals being evaluated for dementia but not yet diagnosed may be included. In addition, we excluded individuals who died during the year or resided in a nursing home because the pattern of continuity and health utilization would likely be different in these groups. Exclusion of beneficiaries with less than 4 outpatient visits, who may have less utilization and spending, also limits the generalizability of our findings. Because our study was cross-sectional, we also cannot make inferences about whether lower COC causes higher health care utilization and spending. Although we accounted for multiple variables that may affect this relationship, there could be reverse causality, with acute events such as hospitalization leading to low continuity. There may also be additional unmeasured confounding factors that influence both continuity and health care utilization. For example, clinical factors leading to specialty referral may also drive utilization. In addition, the COC index is one of several continuity measures; available measures are highly correlated, however, such that using alternatives is unlikely to significantly alter results.^21,50 Finally, we could not consider additional factors related to dementia, such as disease severity, or to continuity, such as relationship duration, trust, and patient perceptions.⁵¹

Lower COC, measured as increased fragmentation of care across clinicians, is associated with greater health care utilization, including hospitalization, ED visits, and testing, and higher health care costs in community-dwelling older adults with a dementia diagnosis. Additional research disentangling the relationship between continuity, types of clinicians seen, and outcomes is indicated as decreasing fragmentation at the clinician level may be a mechanism to reduce hospitalization and unnecessary testing in these older adults at high risk of adverse events. Even within new models of care, emphasis on COC with clinicians may be necessary to improve quality and cost of care for this growing, complex patient population.

Corresponding Author: Halima Amjad, MD, MPH, Johns Hopkins University School of Medicine, 5200 Eastern Ave, Seventh Floor, MFL Center Tower, Baltimore, MD 21224 (hamjad1@jhmi.edu).

Accepted for Publication: May 16, 2016.

Published Online: July 25, 2016. doi:10.1001/jamainternmed.2016.3553.

Author Contributions: Dr Bynum had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Amjad, Austin, Chang, Bynum.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Amjad, Austin, Bynum.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Carmichael, Austin, Chang, Bynum.

Obtained funding: Bynum.

Administrative, technical, or material support: Bynum.

Study supervision: Chang, Bynum.

Conflict of Interest Disclosures: None reported.

Funding/Support: This study was funded by grants from the John A. Hartford Foundation and the National Institute of Aging (P01 AG19783). Dr Amjad was supported by fellowship grants from the Health Resources and Services Administration (D01HP08789) and the Pearl M. Stetler Research Fund.

Role of the Funder/Sponsor: The funding organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.