Care Fragmentation and Emergency Department Use Among Complex Patients With Diabetes

Fragmentation of care to specialists may be associated with higher rates of emergency department visits among patients with chronic illness.

Published Online: January 01, 1970
Constance W. Liu, MD, PhD; Doug Einstadter, MD; and Randall D. Cebul, MD

Objective: To evaluate the association between patterns of fragmented care and emergency department (ED) use among adult patients with diabetes and chronic kidney disease.


Study Design: Observational study in an open healthcare system.


Methods: The study sample included patients with diabetes and chronic kidney disease (mean estimated glomerular filtration rate, 20-60 mL/min) and with an established primary care provider. Dispersion of care was defined by a fragmentation of care index (range, 0-1), with zero reflecting all care in 1 outpatient clinic and 1 reflecting each visit at a different clinic site. We used a negative binomial model to estimate the influence of fragmentation on ED use after adjusting for patient demographic characteristics, insurance, diabetes control, and number of comorbidities; results are reported as incidence rate ratios and associated 95% confidence intervals (CIs). The main outcome measure was the number of ED visits from 2002 to 2003.


Results: Of 3873 patients with diabetes having an established primary care provider, 623 (16.1%) had chronic kidney disease and comprised the final study sample. On average, patients made 19.0 (95% CI, 18.5-20.4) outpatient visits and 1.2 (95% CI, 1.1-1.4) ED visits over the 2-year period. The median fragmentation of care index was 0.48; 14.3% of subjects had a fragmentation of care index of zero. In the adjusted model, a 0.1-U increase in the fragmentation of care index was associated with a 15% increase in the number of ED visits (incidence rate ratio, 1.15; 95% CI, 1.09-1.21).


Conclusions: The posited benefits of specialist referrals among patients with complex diabetes may be partially negated by care fragmentation. Better models for care coordination and stronger evidence of the marginal benefits of referrals are needed.


(Am J Manag Care. 2010;16(6):413-420)

This investigation illustrates that seemingly appropriate referrals to subspecialty providers increased care fragmentation, which was associated with a higher rate of emergency department visits among patients with diabetes and chronic kidney disease.

  • These findings extend previous work that highlights the association of discontinuity of care with worse clinical outcomes, patient dissatisfaction, and provider confusion regarding their roles in care for patients with diabetes.
  • The evidence base for the design of optimal systems of care should be strengthened, with attention to the risks and effectiveness of referrals for patients with diabetes.
Type 2 diabetes mellitus is associated with multiple comorbid illnesses, complications, and hospitalizations thought to be largely preventable by proper patient attention to self-care and by application of best practices of evidence-based care.1 Nonetheless, outcomes of care among patients with diabetes fall far short of expectations. Annual costs of diabetes in the United States are estimated to be more than $130 billion.2 Fewer than half of US patients with diabetes receive recommended health services,3 and a large body of research documents that patients who are less likely to receive needed services are likely to be poorer, less educated, members of racial/ethnic minorities, older, and uninsured.4 As with other ambulatory care–sensitive conditions, however, diabetes outcomes are thought to be improved by better continuity of care, usually defined by a preponderance of visits occurring with the same primary care provider. Emergency department (ED) use, which increased 26% between 1993 and 2003 to about 114 million visits annually, is thought to be a particularly sensitive measure to inadequate outpatient care.5

Providing coordinated continuity of care is challenging in managing the myriad complications that accompany diabetes. Attention to evidence-based guidelines for patients with diabetes even without comorbid conditions appropriately includes annual ophthalmology visits for dilated eye examinations and possibly visits for nutrition counseling and podiatric examinations.6 Patients with diabetes whose glycemia is difficult to control are recommended more aggressive treatment through consultations or comanagement with a diabetes specialist.6 Cardiac and neurologic consultations appropriately may be sought as cardiovascular, cerebrovascular, and neuropathic complications accrue. Professional nephrology organizations have called for earlier referral of patients having diabetes with evidence of renal impairment.7 Accompanying obesity often requires additional referrals for weight management, coexisting sleep apnea, gastroesophageal reflux disease, and arthritis. Approximately 1 in 4 patients with diabetes has coexisting major affective disorders that may require comanagement with a psychiatrist.8-11 In the context of evidence-based guidelines and customary standards of diabetes care, these frequently necessary referrals create the possibility of fragmentation (the dispersion of management of various aspects of a patient’s care to several providers, which may include the patient’s primary care provider or specialists); the effects of frequent referrals on the patient medical home,12 patient-centered quality, and healthcare utilization are largely unknown.

In the present investigation, we examined the effects of care fragmentation on ED use in a particularly complex and vulnerable subgroup of patients having diabetes with kidney impairment. Our principal interest was to determine whether seemingly appropriate referrals to specialists for outpatient screening or consultation for coexisting conditions were offset by the potentially deleterious effects of care fragmentation. Emergency department use was selected as our principal outcome measure, as it is conventionally understood to be “preventable” by optimal ambulatory care of patients with diabetes.13


Subjects and Setting

We selected adult patients having diabetes with chronic kidney disease to identify a homogeneous sample with regard to complexity and need for care. Patients eligible for inclusion in the study included all patients with diabetes older than 18 years and having 2 or more visits in the year before the study period to 1 of 10 primary care group practices within the MetroHealth System, a large public urban provider in Cuyahoga County, Ohio. We required 2 or more visits to the same site to increase the likelihood that these were patients and physicians who were in a continuous relationship, as well as to be consistent with American Diabetes Association recommendations that patients with diabetes have 1 primary care visit every 6 months.14 Data were obtained from electronic medical records over a 2-year study period (2002-2003). Inclusion criteria were as follows: a diabetes mellitus diagnosis in an encounter before the study period (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] code 250.xx), age older than 18 years (determined by age at the beginning of the study period), and the presence of chronic kidney disease, defined as a 2-year mean glomerular filtration rate (GFR) between 20 and 60 mL/min. The GFR was calculated using the Mater Medical Research Institute equation, which relies on the variables of sex, race/ethnicity, and recorded creatinine levels.15 Patients were excluded from the study if they received their continuity care primarily from resident physicians or if their mean GFR was less than 20 mL/min.

Main Outcome Measure

The primary outcome variable was the number of ED visits within the MetroHealth System made by patients over the 2-year study period. Any visit labeled as an ED visit in the encounter data contributed toward the count.

Independent Variables

We operationalized fragmentation of patient care using a fragmentation of care index (FCI). The FCI is based on a modified version of the previously validated continuity of care index.16 The continuity of care index, first described by Bice and Boxerman,17 is based on the number of different providers visited, the proportion of attended visits to each of those providers, and the total number of visits. Our FCI measure used clinics as the unit of measurement, rather than individual providers, and further modified the continuity of care index as follows to compute the FCI:

where n indicates the total number of visits; nk, the total number of visits to clinic k; and k, the number of clinics. The FCI can range from 0 (all visits to the same clinic) to 1 (each visit takes place at a different clinic). Visits  classified into 1 of the following categories, based on plausible referral specialties for diabetes-related comorbidities or monitoring, were included in the FCI: primary care, cardiology, neurology, endocrinology, ophthalmology, nephrology, urology, pulmonary medicine, podiatry, gastroenterology, and psychiatry.

A practical demonstration of the behavior of the FCI may help in its interpretation. Assume a patient makes 12 visits to the primary care clinic, 1 visit to the ophthalmology clinic, and 2 visits to the podiatry clinic. Her FCI would be 0.36. If the patient’s glucose remains poorly controlled, the referring primary care physician might see her more frequently or might decide instead to refer her to the endocrinology clinic. If the patient makes an additional 4 visits to the primary care clinic, the FCI decreases 0.07, from 0.36 to 0.29. If the additional 4 visits instead are to the endocrinology clinic, the dispersion of visits to an additional provider would increase the FCI by 0.21, from 0.36 to 0.57.


Potentially confounding demographic variables used in the analysis included race/ethnicity (black, white, Hispanic, or other), age at the beginning of the study period, sex, and insurance status (commercial, Medicaid, Medicare, or uninsured). Potential confounding clinical variables that were adjusted for included severity of diabetes (defined as the mean glycosylated hemoglobin level over the 2-year study period) and a count of comorbidities.18 Comorbid illnesses and their associated ICD-9-CM codes included arthritis (codes 711, 712, 714-716, 720, and 726), coronary artery disease (code 414), cancer (codes 140-239), congestive heart failure (code 428), chronic obstructive pulmonary disease (codes 490-496), hypertension (codes 401-405), liver disease (code 751), stroke (codes 430-438), and psychiatric disease (codes 295-301, 308, 309, and 311). Diabetes and renal disease were not included in the comorbidity count.

Statistical Analysis

Frequencies, means, and standard deviations were used to describe the sample population. Negative binomial regression was used to describe the association between the FCI and the number of ED visits, after adjustment for potential confounders. A goodness-of-fit test verified that this model was the most appropriate to the nonrandom overdispersed distribution of ED visit count. The multivariate model was built in a forward stepwise fashion, adjusting for demographic and clinical confounders. Age, sex, race/ethnicity, insurance status, glycosylated hemoglobin level, and number of comorbidities were entered as adjusting variables. Results of the model represent the change in ED visits corresponding to a 0.1-U change in the FCI. For all statistical tests, P <.05 was considered statistically significant. We used commercially available statistical software (STATA, version 7.0 for Windows; StataCorp LP, College Station, TX). The protocol was reviewed and approved by the MetroHealth Medical Center Institutional Review Board.


PDF is available on the last page.
Adult ADHD Compendium
COPD Compendium
Dermatology Compendium
Diabetes Compendium
GI Compendium
Hematology Compendium
Immuno-oncology Compendium
Lipids Compendium
MACRA Compendium
Oncology Compendium
Pain Compendium
Reimbursement Compendium
Rheumatoid Arthritis Compendium
Know Your News
HF Compendium
Managed Care PODCAST