Zhen-qiang Ma, MD, MPH, MS

To evaluate the associations between potentially avoidable diabetes-related emergency/urgent hospital admissions and different health insurance status (ie, Medicaid, Medicare,

	uninsured, private), along with other characteristics including sociodemographic status (age, race/ethnicity, gender, region), hospitalization status (previous hospitalizations, weekend admissions), and health status (complications, comorbidities), among patients with type 2 diabetes mellitus (T2DM).

 The 2011 data set of all inpatient discharge records with a primary diagnosis of T2DM from all hospitals in Pennsylvania were included in the analyses.

Multivariable logistic regression modeling with diabetes-related emergency/urgent hospitalizations as the dependent outcome variable and health insurance status as the main exposure independent variable, adjusting for age, race/ethnicity, gender, region, previous hospitalizations, weekend admissions, complications, and comorbidity. Hosmer and Lemeshow

	goodness-of-fit test was used for logistic model fit analysis.

Nearly 91% of 17,097 potentially avoidable diabetes-related hospitalizations were emergency/urgent admissions for T2DM patients in Pennsylvania during 2011. Uninsured and Medicaidinsured patients were 2.1 (adjusted odds ratio [AOR], 2.11; 95% CI, 1.23-3.61) and 1.8 (AOR, 1.78; 95% CI, 1.44-2.20) times more likely than privately insured patients, respectively, to be admitted through emergency/urgent admissions. There was no statistically significant difference in emergency/urgent admissions between Medicaid and uninsured (AOR, 0.85; 95% CI, 0.49-1.47).

Medicaid-insured T2DM patients, like the uninsured, are more likely to be hospitalized through emergency/urgent admissions. The presumption that insured individuals with diabetes are more likely than the uninsured to manage and control the progression of their condition, and receive care in the right setting, is not supported for those with Medicaid coverage.

	Multivariate analyses of medical insurance status affecting diabetes-related emergency/urgent hospitalization charges are sparse, but essential when making managed care decisions regarding health coverage expansion.

		Medicaid-insured type 2 diabetes mellitus (T2DM) patients, just like the uninsured, are more likely to be hospitalized through emergency/urgent admissions.

		The presumption that insured individuals with T2DM are more likely than the uninsured to manage and control the progression of their condition, and receive care in the right setting, is not supported for those with Medicaid coverage.

		Further research is needed to assess the services provided by Medicaid and the health insurance exchanges that promote compliance with self-care and management recommendations, and improve access to and utilization of appropriate ambulatory/primary care services.

Diabetes and diabetes-related comorbidity are important public health problems due to the substantial disease and cost burdens.

 Approximately 22.3 million Americans, or 7% of the US population, were diagnosed with diabetes in 2012; individuals with diabetes account for more than 20% of US healthcare dollars spent, with hospital inpatient care incurring 43% of the $176 billion total direct diabetes medical cost.

 Because hospitalizations account for the largest share of total healthcare costs, the Agency for Healthcare Research and Quality (AHRQ) developed Prevention Quality Indicators (PQIs) to identify hospital admissions that were potentially avoidable with timely access to primary care and appropriate disease management.

 Using AHRQ’s PQIs as the recognized metric, an estimated 12.4% of the hospitalizations in Pennsylvania are potentially avoidable, and of that 12.4%, 1 in 8 are diabetes-related,

 despite 96% of the diabetes-related hospitalizations being preventable.

Hospitalization costs related to poor management of diabetes are major contributors to the healthcare costs of those with diabetes. This is especially important when considering Medicaid expansion under the Affordable Care Act (ACA), because Medicaid and Medicare are the primary payers for 49% of the diabetes-related emergency/urgent hospital admissions.

 Studies assess factors related to avoidable hospitalizations as indicators of access to primary healthcare and corresponding healthcare costs.

 Health insurance is a key determinant of access to care and health outcomes,

 such that health insurance status can significantly impact effective management and control of diabetes. For instance, both uninsured and Medicaid-insured individuals have less favorable health outcomes than those who are privately insured and/or Medicare-insured,

 In addition to health insurance status, having a previous hospital visit or readmission is typically assessed in studies of potentially avoidable hospitalizations.

However, not all avoidable hospitalizations are the same. In particular, whether a hospital visit is an emergency admission or non-emergency admission for ambulatory care—sensitive conditions (eg, diabetes) is an indicator of process and coordination of care.

 That is, unplanned emergency/urgent hospital admissions are an indicator of poorly managed acute diabetes conditions, whereas elective or planned admissions generally suggest that patients received diabetes management care in the right setting such that their conditions permitted adequate time to schedule the services. Due to the paucity of scientific evidence, studies are especially needed to gain a better understanding of the effect of health insurance status on emergency/urgent admissions, because 94% of potentially avoidable diabetes-related hospitalizations fall into this category.

To better understand the factors influencing the potentially avoidable emergency/urgent diabetes-related hospitalizations, the objective of this study is to quantify the independent association between potentially avoidable diabetes-related emergency/urgent hospital admissions and different health insurance status (ie, Medicaid, Medicare, uninsured, private), along with other characteristics including sociodemographic status (ie, age, race/ethnicity, gender, region), hospitalization status (ie, previous hospitalizations, weekend admissions), and health status (ie, complications, comorbidities), among those with the most common type of diabetes, type 2 diabetes mellitus (T2DM).

The Pennsylvania Health Care Cost Containment Council (PHC4)—an independent state agency that collects approximately 1.9 million inpatient discharge records annually from all hospitals in Pennsylvania&mdash;provided the 2011 data set of inpatient discharge records from all the state’s hospitals, which was then analyzed. Potentially avoidable diabetes-related hospitalizations were defined based on AHRQ’s PQIs that are used with inpatient hospital discharge data to identify conditions that should be preventable with proper primary care.

 Potentially avoidable T2DM-related hospitalizations were defined as all hospitalizations with a primary discharge diagnosis of T2DM (

International Classification of Diseases, Ninth Revision, Clinical Modification

] codes of 250.00-250.92 with a last digit of 0 or 2).

Description of outcome (dependent variable).

 The outcome, emergency/urgent hospital admission, was dichotomized (yes/no) based on the admission type that defined the hospital admission’s level of urgency. An emergency/urgent admission was defined as the patient requiring immediate medical intervention as a result of a severe, life-threatening, or potentially disabling condition; or the patient required immediate attention for the care and treatment of a physical or mental disorder. A non-emergency/urgent admission was defined as an elective/planned admission, with the patient’s condition permitting adequate time to schedule the services.

Description of main exposure (independent variable).

 Health insurance status was based on the primary payer type and categorized as Medicaid, Medicare, uninsured, and private.

Description of Other Explanatory Factors (Independent Variables)

The 4 sociodemographic characteristics that were assessed include age, race/ethnicity, gender, and region: 1) age: the patient’s age in years at the time of hospitalization, categorized as <20, 20-44, 45-64, or ≥65

; 2) race/ethnicity: patients were categorized as non-Hispanic white (NHW), non-Hispanic black, non-Hispanic other race, or Hispanic based on the combination of race and ethnicity information; 3) gender: if gender information was missing, it was imputed from prior (2000-2011) records; 4) region: facility region code was categorized as Philadelphia (includes 4 surrounding counties), Pittsburgh (includes Pittsburgh city, Allegheny County and 7 surrounding counties), and the rest of  the state (other 53 counties).

The 4 hospitalization and health-related statuses that were assessed include previous hospitalization, weekend admission, complications, and comorbidities: 1) previous hospitalization: categorized as “yes” or “no” and defined as any prior hospital admission with primary diagnosis of diabetes from 2000 to 2011; 2) weekend admission: categorized as “yes” or “no” based on the date of admission; 3) complications: defined based on the AHRQ’s PQI #1 Diabetes Short-Term Complications Admission Rate, and PQI #3 Diabetes Long-Term Complications Rate, along with PQI #14 Uncontrolled Diabetes Admission Rates.

 Complications were derived from the fourth digit of the primary diagnosis 

 codes and categorized as: a) no complication; b) acute: coma, hyperosmolarity, ketoacidosis; c) chronic: renal manifestations, neurological manifestations, ophthalmic manifestations, peripheral circulatory disorders, and other complications.

 Thus, complications are a proxy measure for the progression of diabetes; 4) comorbidities as defined by the modified Charlson comorbidity index (CCI): the CCI predicts 10-year mortality based on weighted scores ranging from 0 to 35 for 17 conditions.

 All 8 secondary diagnoses were considered in the calculation of the CCI. Diabetes was excluded from CCI calculation so as not to double-count the impact of diabetes, because all assessed hospitalizations were diabetes-related. The CCI distribution ranged from 0 to 12, with 24% having a score of 3 or more. CCI categories were defined as 0, 1, 2, and 3 or more in order to keep the distribution (or data) balanced.

First, descriptive analyses were performed to report the overall and characteristic-specific (independent variables) proportion of potentially avoidable diabetes-related hospitalizations that were emergency/urgent admissions for Pennsylvanians in 2011. This was followed by the corresponding unadjusted association or crude odds ratio (COR) between emergency/urgent hospitalizations and insurance status, sociodemographic status, hospitalization, and health status.

Multivariable logistic regression modeling

The independent association was calculated as the adjusted odds ratio (AOR) for potentially avoidable diabetes-related emergency/urgent hospitalizations and health insurance status, while simultaneously taking into account sociodemographic status, hospitalization status, and health status. Hosmer and Lemeshow’s goodness-of-fit test was used for logistic model fit analysis. The observed event rates matched expected event rates in all subgroups of the population in the final model.

 Among 17,097 potentially avoidable diabetes-related hospitalizations in Pennsylvania in 2011, 15,538 (90.9%) were emergency/urgent admissions. When considering only a single characteristic, the following associations with having potentially avoidable diabetes-related emergency/urgent admissions were quantified: 1) health insurance status: uninsured and Medicaid-insured patients were 3.1 and 2.4 times more likely, respectively, than privately insured patients; and Medicare-insured patients were approximately one-eighth less likely; 2) sociodemographics: younger patients were 1.3 to 2.7 times more likely than older patients aged ≥65 years, and minority patients were 2.2 to 2.9 times more likely than NHW patients. Hospitalizations in Philadelphia and Pittsburgh were approximately 3.1 times and 1.2 times, more likely, respectively, than admissions in the rest of the state; 3) hospitalization status: weekend admissions were 4.7 times more likely than weekday admissions; and first hospitalizations were 1.4 times more likely than those with a previous diabetes-related hospitalization(s); 4) health status: those with acute complications were 3.4 times more likely, and those with chronic complications were 2.6 times less likely, than those without complications, respectively. Patients with comorbidities were one-half to one-third as likely as patients with no comorbidity.

Multivariable logistic regression modeling 

After simultaneously taking into account all 9 characteristics listed in Table 2, the following independent associations with having potentially avoidable diabetes-related emergency/urgent admissions were quantified: 1) health insurance status: uninsured and Medicaid-insured patients were 2.1 and 1.8 times more likely, respectively, than privately insured patients. There was no statistical difference between Medicaid-insured and uninsured (AOR, 0.85; 95% CI, 0.49-1.47 using uninsured as the reference group in the model); or between Medicare-insured and privately insured; 2) sociodemographics: younger patients aged <20 years were approximately half as likely as patients aged ≥65 years, and minority patients were 1.5 times more likely than NHW patients. Hospitalizations in Philadelphia and Pittsburgh were 2.6 times and 1.2 times, respectively, more likely than admissions in the rest of state; 3) hospitalization status: first hospitalizations were 1.4 times more likely than those for patients with a previous diabetes-related hospitalization(s), and weekend admissions were 4.3 times more likely than weekday admissions; 4) health status: those with acute complications were 3.1 times more likely, and those with chronic complications were half as likely, as those with no complications. Patients with comorbidities were approximately one-fifth as likely as patients with no comorbidity.

Hospitalization costs related to poor management of diabetes are major contributors to the healthcare costs of those with diabetes. This study was designed to investigate the factors influencing the substantial disease and cost burden of potentially avoidable and expensive diabetes-related emergency/urgent hospital admissions, indicative of poor diabetes management and control. It is noteworthy that most previous studies of potentially avoidable hospitalizations did not differentiate between emergency and non-emergency admissions, resulting in misclassification bias, because it is the emergency admissions that are most amenable to interventions that improve ambulatory disease management and control. By addressing this misclassification bias, our study provides relevant scientific evidence regarding factors (eg, health insurance status, sociodemographic characteristics, hospitalization, and health status) related to the 94% of potentially avoidable diabetes-related hospitalizations that are emergency/urgent admissions.

 Analyses such as these, focusing on ambulatory care—sensitive conditions, illustrate the opportunity to reduce potentially avoidable hospital admissions and redirect resources to other types of healthcare services.

Furthermore, as states consider Medicaid expansion, they should be cognizant of the issues surrounding Medicaid recipients’ access to care, utilization of services, and compliance with self-management recommendations, not just health insurance status. As indicated in our analysis, Medicaid-insured diabetes-related hospitalizations for those with T2DM were similar to those of the uninsured. Their admissions were more likely to be emergency/urgent admissions than were those of the privately insured or Medicare-insured, after simultaneously taking into account other factors. This key finding—that having Medicaid insurance is not much different from being uninsured&mdash;necessitates further investigation, especially when publicly provided or publicly subsidized health insurance coverage for all is being considered. That is, when attempting to understand the lack of routine healthcare utilization, and to understand issues related to better management and control of diabetes, ultimately, it is important to consider some things other than simply having insurance. For example, financial constraints most likely influence healthcare-seeking behaviors of both uninsured and Medicaid recipients. In addition, lapses in Medicaid coverage have been associated with increased inpatient and emergency department (ED) services in the post lapse period.

 Hence, one should be cognizant that previously uninsured individuals who are later covered by publicly provided or subsidized health insurance may continue to use healthcare services differently from those who were privately insured or Medicare-insured, and may continue to use emergency/urgent care rather than obtain care through primary care providers.

More specifically, simply having insurance coverage most likely will not resolve the issue of increasing healthcare costs because of the multitude of intertwined factors related to preventive care, having a medical home, committing to self-management (eg, through diet, exercise, medication adherence), access to care, and utilizing health services in the right setting (eg, seeking primary care when appropriate, being able to make and keep appointments, being able to get time off from work, taking into account a possibly long wait time for an appointment). These and other factors may mean that poor and working-poor Medicaid-insured and uninsured individuals may not be able to optimally seek and utilize preventive care services in an ambulatory care setting. Instead, they may end up in the ED, being admitted to the hospital for potentially avoidable diabetes-related conditions.

For example, it is commendable to focus on person-centered care through a medical home model,

 and to create Teaching Health Center Graduate Medical Education programs in ambulatory care centers in underserved areas through the ACA.

 However, these efforts will not ultimately improve health outcomes and decrease healthcare costs if patients with chronic conditions such as T2DM do not have regular preventive visits in an ambulatory care setting at the right time, to manage and control the progression of their conditions.

 To give another example, quality improvement approaches that focus on educational interventions have reportedly improved medication adherence and decreased acute care services (ie, hospitalization, emergency visits).

 reports that the effect of quality improvement programs for managing diabetes, which include financial incentives to directly reward primary care physicians, has not been clearly established.

These seemingly inconsistent findings suggest that while changes in clinical care pathways seem to make sense, the translation of knowledge to behavioral change is complex, and there is no easy solution. Additional research that builds upon the currently available science—such as applying a version of the “5 As” approach that is a cornerstone of some smoking cessation plans

—could provide some helpful insight when counseling patients in how to improve their diabetes self-management behaviors. While the desire to change behavior is necessary, it is not sufficient&mdash;one also needs concrete knowledge regarding appropriate self-care/management.

Some additional barriers to behavioral change, especially among the poor and working poor, include food insecurity, which can impede the dietary component of diabetes self-management,

 and health insurance status, which has been associated with access to appointments, timely appointments, and affordability of ambulatory care. When out-of-pocket payments were limited to less than $50, only 14% of uninsured people in one study were able to schedule an appointment, whereas without this payment limitation 73% of uninsured, 71% of privately insured, 62% of Medicare-insured, and 37% of Medicaid-insured people were able to do so. These percentages are further reduced for an appointment within 7 days with a $50 out-of-pocket cost.

 Intervention strategies that modify the primary care physician financial incentive program,

such as financial incentives to individual patients, may be worth consideration, especially for uninsured or Medicaid-insured individuals who have substantial financial constraints.

This study of the association between potentially avoidable diabetes-related emergency/urgent hospitalizations and health insurance status among those with T2DM had several advantages. First of all, of interest to insurance providers (ie, private, Medicaid, Medicare), the more rigorous multivariable model has stronger data-driven policy implications for disease prevention, management, and control than simple descriptive and/or unadjusted crude associations. More specifically, the diabetes-related hospitalizations with acute complications were over 3 times more likely to have emergency/urgent hospital admissions than those without complications. Thus, considerable cost savings are feasible when informed policies impede the development and progression of T2DM, and ultimately prevent or substantially decrease emergency/urgent hospital admissions.

Secondly, comorbidities were also taken into account, based on the weighted composite CCI. Diabetes-related hospitalizations were less likely to be emergency/urgent admissions for those with T2DM and comorbidity. This apparent paradox may be partially explained by the speculation that those with 1 or more comorbidities may see multiple providers/specialists; cumulatively, those providers may do a better job at monitoring and controlling diabetes, so the patient is less likely to have an emergency/urgent hospital admission. Alternatively, patients who have T2DM alone may not believe it’s as important to manage and control their condition as do those with multiple comorbidities. They may be lackadaisical about scheduling primary care appointments to manage diabetes alone, with the result being the need for emergency/urgent acute care services requiring hospitalization. Further research of the role(s) of specific components of the CCI measure may provide relevant insight when developing potential recommendations and policy changes that address the utilization of primary care versus ED services.

Thirdly, the fact that those who had previous diabetes-related hospitalization(s) were less likely to have a subsequent emergency/urgent admission suggests that the hospital experience may have changed the patterns of care for these patients. Additional insight may be gleaned from investigations of the effectiveness of improved clinical pathways that focus on multifaceted patient-oriented interventions such as education and compliance with treatment,

 financial incentives to primary care physicians,

 and post discharge programs that assess compliance with insulin therapy after hospitalization.

We purposefully restricted our analyses to the more common type 2 form of diabetes because types 1 and 2 are very different diseases medically. We selected diabetes hospitalization with primary diagnosis of diabetes only in our analyses. This is consistent with standard definitions used by AHRQ’s PQI and Healthcare Cost and Utilization Project to identify conditions that should be preventable with proper primary care.

 Hence, this approach eliminated the potential misclassification bias that is present when admissions with a secondary diagnosis of diabetes, such as hospitalizations for joint replacement surgery or for trauma related to a car accident, are inappropriately considered to be potentially avoidable diabetes-related hospitalizations. It is also noteworthy that among patients with T2DM, diabetes-related hospitalizations in the Philadelphia region were significantly more likely to have expensive emergency/urgent admissions than in Pittsburgh or the rest of the state. This finding is consistent with the report that patients who present to the ED were more likely to be admitted to the hospital in metropolitan areas (13.6%) than in rural areas (8.3%).

We believe the aforementioned strengths of this study far outweigh the following limitations. It would have been informative if data were available on income or education; that might partially explain the results that Medicaid-insured and uninsured diabetes-related hospitalizations were more likely to have begun with emergency/urgent admissions.

 Medicaid-insured and uninsured populations were similar in that both were more likely to have emergency/urgent hospitalizations than those privately insured or Medicare-insured; this suggests that having insurance coverage is not sufficient when considering associated costly health behaviors. As discussed earlier, different financial constraints may affect the poor and working poor, to limit their access to, and utilization of, primary care services that would help manage and control the progression of diabetes.

The presumption that insured individuals with T2DM are more likely than the uninsured to manage and control the progression of their condition, and to receive the needed care at the right time in the right setting, is not supported for those insured by Medicaid. These data have potential implications for states considering Medicaid expansion under the ACA. Further research is needed to assess the services provided by Medicaid and the health insurance exchanges that promote compliance with self-care and management recommendations, and to determine if access to care and utilization of appropriate ambulatory/primary care services is improved.

The authors acknowledge the support, review, and insightful comments provided by Stephen M. Ostroff, MD, director of the Bureau of Epidemiology at the Pennsylvania Department of Health at the time this study was initiated. Support from Judith Good at Pennsylvania Health Care Cost Containment Council for extracting the hospital discharge data for this analysis is also acknowledged. The Pennsylvania Health Care Cost Containment Council (PHC4) is an independent state agency responsible for addressing the problem of escalating health costs, ensuring the quality of healthcare, and increasing access to healthcare for all citizens. While PHC4 has provided data for this study, PHC4 specifically disclaims responsibility for any analyses, interpretations or conclusions.

College of Medicine, Central Michigan University (MAF), Mount Pleasant, MI; Pennsylvania Department of Health, Bureau of Epidemiology (ZM), Harrisburg, PA.

This study was funded by the CDC Coordinated Chronic Disease Prevention grant, CDC-RFA-DP09-9010301PPHF11 and National Association of Chronic Disease Directors Epidemiology Capacity Building grant, NACDD-0612012.

The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Concept and design (MAF, ZM); acquisition of data (ZM); analysis and interpretation of data (MAF, ZM); drafting of the manuscript (MAF, ZM); critical revision of the manuscript for important intellectual content (MAF, ZM); statistical analysis (ZM).

Zhen-qiang Ma, MD, MPH, MS, Pennsylvania Department of Health, Bureau of Epidemiology, 625 Forster St, Harrisburg, PA 17120. E-mail: 

1. American Diabetes Association. Economic costs of diabetes in the U.S. in 2012. 

2. Mayes R, Oliver TR. Chronic disease and the shifting focus of public health: is prevention still a political lightweight? 

Articles

Medicaid-insured type 2 diabetes mellitus patients, just like the uninsured, are more likely to be hospitalized through emergency/urgent admissions.

Medicaid-Insured and Uninsured Were More Likely to Have Diabetes Emergency/Urgent Admissions

To evaluate the characteristics driving higher diabetes-related hospitalization charges.

Hospital discharge data on 216,858 hospitalizations from 2001 to 2011 with primary discharge diagnosis of diabetes were linked to the Pennsylvania death registry. Multiple linear regression analysis was used to evaluate the association between inpatient hospitalization charges and complications, sociodemographic status, comorbidities, readmission, length of stay, admission type, region, procedures, payer type, and hospitalization misadventures.

Diabetes-related adjusted hospitalization charges were higher for those with long-term complications [renal manifestations (75%), peripheral circulatory disorders (38%), neurological manifestations (30%), ophthalmic manifestations (16%)], acute complications [ketoacidosis (48%), hyperosmolarity (41%), coma (40%)], amputations (91%) or other medical procedure(s) (70%), emergency/urgent admissions (7%), comorbidity (3% per modified Charlson Comorbidity Index score item), medical misadventure (20%), different regions [Philadelphia (135%), and Pittsburgh (8%)], and for minorities [non-Hispanic black (12%), Hispanic (21%), and non-Hispanic other (14%)]. Readmission adjusted charges were not different from the initial admission charges.

	• Diabetes complications, medical procedures, region, race/ethnicity, and medical misadventures are major drivers of diabetes-related hospitalization charges. Age, admission type, payer type, and comorbidities are also independent drivers of these charges.

	• Evidence-based clinical pathways, policy changes that reduce medical misadventures and assess discharge readiness, and improving quality of follow-up care after discharge should reduce unnecessary charges and readmissions.

	• Future research is needed to investigate why hospital charges differ by region, payer type, and race/ethnicity, after simultaneously taking into account 16 contributing factors.

To reduce diabetes-related hospitalizations and curb hospitalization charges, public health and healthcare policy makers should be cognizant of high-impact drivers: diabetes-related complications, unnecessary procedures, race/ethnicity, and region. Inpatient are should focus on preventing unnecessary readmissions and misadventures. These findings are timely as the Affordable Care Act reduces Medicare payments to hospitals with high readmission rates, and as states consider Medicaid expansion.Multiple linear regression analyses of factors driving diabetes-related hospitalization charges are sparse, but essential when making managed care decisions regarding healthcare reform.

Diabetes is an important public health problem, affecting an estimated 25.8 million Americans (8.3% of the US population) as of 2010.

 In Pennsylvania, our area of focus, the prevalence of adults diagnosed with diabetes has considerably increased, almost doubling from an estimated 500,000 (5.5%) in 1994 to 941,000 (9.7%) in 2010.

 In addition to the morbidity and mortality burden, the economic burden of diabetes in the United States was estimated as $174 billion in 2007, including direct medical costs of $116 billion and indirect disability, work loss, and premature mortality costs of $58 billion.

 The average medical expenditures for those with diagnosed diabetes are 2.3 times higher than the expenditures for those without diabetes.

Hospitalization costs are a major contribution to the healthcare costs of those with diabetes,

 exacerbated by the economic burden of poor management. Data regarding the characteristics that drive higher hospitalization charges are sparse, but they are crucial; the Affordable Care Act (ACA) began reducing Medicare payments to hospitals with high readmissions rates during Fiscal Year 2013.

 Comprehensive evaluations that quantify the multiple patient and hospital factors that independently drive the variation in diabetes hospital charges are scarce. Studies designed to address this gap in knowledge can facilitate the rendering of more informed data-driven decisions. The data reported herein add to the scientific evidence by providing such adjusted diabetes hospitalization charges. We simultaneously consider complications, sociodemographic status, comorbidities, readmission, length of stay (LOS), admission type, region, procedures, payer type, and medical misadventures.

By investigating this question—“What patient and discharge characteristics independently drive hospitalization charges related to poor diabetes management in Pennsylvania?”&mdash;the study’s objective was to provide much-needed relevant data for states as the ACA reduces Medicare payments to hospitals with high readmissions, and as states consider expansion of Medicaid. The analytic approach simultaneously takes into account all patient and hospitalization factors associated with hospitalization charges: acute and chronic complications, comorbidity, sociodemographic status, readmission, LOS, admission type, region, procedures, payer type, and medical misadventure. Such an approach allows healthcare policy makers and decision makers to develop evidence-based approaches that target the high-impact drivers of hospitalization charges.

A diabetes-related hospitalization was defined as a patient hospitalized with a primary diagnosis of diabetes (all 250 codes in the 

 [ICD-9-CM]) at discharge in the Pennsylvania Health Care Cost Containment Council (PHC4) data set during the period of 2001 to 2011, with a total hospitalization room charge greater than $0 and an LOS less than 90 days. PHC4 data were linked to Pennsylvania vital statistics death certificate data by PHC4, using patient identity match algorithms to get death information.

Description of Hospitalization Charge Outcome (dependent variable)

Hospitalization charge was defined as the total charge related to the hospital stay with a primary diagnosis of diabetes excluding professional fees. This includes room and board charges, ancillary charges, drug charges, equipment charges, specialty charges, and miscellaneous charges.

Description of Patient and Discharge Characteristics (independent variables)

 Derived from the fourth digit of the primary and all 8 secondary diagnoses ICD-9-CM codes, and were grouped into the following categories: no complication, coma, hyperosmolarity, ketoacidosis, renal manifestations, neurological manifestations, ophthalmic manifestations, peripheral circulatory disorders, other complications, and 2 or more complications. These are based on the Agency for Healthcare Research Quality (AHRQ) Prevention Quality Indicators (PQI) technical specifications, version 4.4.

 Inpatient hospitalization year at date of admission, from 2001 to 2011.

Patient age in years at time of hospitalization. Categorized as <20, 20 to 34, 35 to 44, 45 to 64, ≥65 years.

 Patients were categorized as non-Hispanic white (NHW), non-Hispanic black (NHB), non-Hispanic other race (NHO), and Hispanic. If race/ethnicity was missing for a record, it was inputted from a previous or subsequent record in the data set.

 If gender was missing for a record, it was inputted from another record in the dataset.

 Type 1 diabetes (T1DM) or type 2 diabetes (T2DM) was defined based on the last digit of the primary ICD-9-CM diagnosis code; an even number is T2DM, an odd number is T1DM.

 Readmission days were calculated from the previous hospitalization discharge date with a primary discharge diagnosis of diabetes. Readmissions were categorized as 1 to 7 days, 8 to 14 days, 15 to 30 days, after 30 days, and first admission. Year 2000 data were consulted to ensure that all hospitalizations could be assessed as a new hospitalization or readmission for a minimum of 1 year.

Patients with discharge status of “expired” or having the same discharge date and death date on their death certificate were defined as “death at discharge.” Based on the difference between death date and the discharge date, others were categorized as either death between 1 and 7 days after discharge, and alive after 7 days of discharge.

 Categorized as commercial, Medicaid, Medicare, uninsured, and other.

 Defined as the hospital admission’s level of urgency; 1) Emergency department (ED)/urgent admissions were defined as such when the patient required immediate medical intervention as a result of severe, life-threatening, or potentially disabling conditions; urgent admissions were noted when the patient required immediate attention for the care and treatment of a physical or mental disorder; 2) Elective admissions were defined as cases in which the patient’s condition permitted adequate time to schedule the services.

 Facility region was categorized as Philadelphia (includes the surrounding 4 counties of Chester, Montgomery, Bucks and Delaware); Pittsburgh (includes Pittsburgh city, Allegheny County, and the 7 surrounding counties of Armstrong, Beaver, Butler, Fayette, Greene, Washington, Westmoreland), or other (the rest of the state of Pennsylvania).

 Defined based on the primary and 5 secondary procedure codes, then further categorized as no procedure, amputation, and procedures other than amputation.

Defined as the patient having external causes of injury and poisoning based on the presence of one or more codes between E870 and E879. These include misadventures to patients during surgical and medical care (E870-E876), and surgical and medical procedures as the cause of abnormal reaction of patient or later complication, without mention of misadventure at the time of procedure (E878-E879).

 The number of hospitalization days with a maximum of 90 days; LOS of 0 was recoded to 0.5 to avoid counting partial-day admissions as 0.

Modified Charlson Comorbidity Index (CCI):

 Predicts 10-year mortality based on weighted scores ranging from 0 to 35 for 19 conditions.

 All 8 secondary diagnoses were considered in the calculation of the CCI. Diabetes was excluded from the weighted score so the impact of diabetes would not be double-counted; all the hospitalizations we assessed were related to diabetes.

Total number of hospitalizations and average hospitalization charges were calculated by year, complication, age group, race/ethnicity, gender, diabetes type, readmission status, mortality status, primary payer type, admission type, admission day of the week, region, procedure(s), and medical misadventure. Average LOS and modified CCI were also calculated.

Hospitalization charge was the dependent variable and the following characteristics were categorical independent variables: complication, age, race/ethnicity, gender, diabetes type, readmission status, mortality status, primary payer type, admission type, admission day, region, procedures, and medical misadventure. Year, LOS, and modified CCI were continuous independent variables. Initial model fitting diagnosis revealed a funnel-shaped residual plot, which indicated nonconstant variance. Natural log transformation of the hospitalization charge was used to correct heteroskedasticity.

 Multicollinearity was examined in the final transformed hospitalization charge model for year, LOS, modified CCI, and medical misadventure. All variance inflation factors and condition numbers were less than 2 and 5, respectively, indicating multicollinearity was not an issue. SAS version 9.3 (SAS Institute, Cary, North Carolina) was used for all analyses. A P value of <.05 was considered statistically significant.

To clearly illustrate the merit of our analytic approach, we will first describe the average unadjusted hospitalization charges for 216,858 diabetes-related hospitalizations from 2001 to 2011, depicted in 

. This was followed by the more informative adjusted hospitalization charges that simultaneously take into account all 16 factors in the multiple linear regression model depicted in 

The overall average unadjusted hospital charges consistently increased every year without adjustment for inflation from $19,167 in 2001 to $41,246 in 2011. On average there was an 8.6% increase in adjusted hospitalization charges every year.

 Average unadjusted hospitalization charges were 158% higher for those with any type of complication ($34,577) than for those without complications ($13,417), ranging from 73% higher for those with ophthalmic manifestations to 431% higher for those with renal manifestations, than for those who did not have complications. However, after adjusting for the other 15 factors the increase in charges for those with complications reduced substantially, ranging from 16% higher for those with ophthalmic complications or other complications, to 75% higher for those with renal manifestations.

 The average unadjusted hospitalization charge for patients aged 45 to 64 years ($37,572) was 166% higher than for those younger than 20 years ($14,147) and 18% higher than for those 65 years and older ($31,854). Adjusted hospitalization charges for patients aged 45 to 64 years was 17% higher than for patients younger than 20 years, a significant reduction compared with the unadjusted difference of 166%.

Average unadjusted hospitalization charges to NHBs ($43,117) were nearly 60% higher than the charges to NHWs ($27,156), but this difference reduced to 12% after adjusting for the other factors. The average unadjusted hospitalization charge to Hispanics ($40,943) was 51% higher than to NHWs, but reduced to 21% higher after adjusting for the other factors.

 Men’s average unadjusted hospitalization charges were $4331, 15% more than for women, but there was no gender difference in adjusted hospitalization charges (P = .0520).

 Average unadjusted hospitalization charge for those with T1DM was 19% less than for those with T2DM, but was 5% higher after adjusting for the other factors.

Average unadjusted hospitalization charge for readmission ranged from 27% to 35% higher than for the first admission, but was no longer significantly different after adjusting for the other factors. In other words, adjusted hospitalization charges for readmissions were about the same as the initial hospitalization charges.

 Average unadjusted hospitalization charges for patients who died in the hospital or on the same date that they were discharged were 202% higher, and for those who died within a week after discharge were 55% higher, than for those who were alive after 7 days of discharge. After adjusting for the other factors, the difference reduced to 6% for death at discharge, and was no longer significant for those who died within a week after discharge.

 Average unadjusted hospitalization charges to those with commercial, Medicare, Medicaid, and other types of medical insurance were, respectively, 32%, 62%, 48%, and 20% higher than the charges to uninsured. This reduced to 11% for commercially insured, 13% for Medicare, 10% for Medicaid, and 8% for other types of insurances after adjusting for other factors.

 Average unadjusted hospitalization charges to those who were ED or urgent admissions were 10% less than those with elective admissions, but the adjusted charges were 7% higher for ED/urgent admissions.

 Average unadjusted hospitalization charges for weekday admissions ($32,196) were 9% higher than for weekend admissions ($29,677), but the adjusted charges were 3% higher for weekend admissions.

 Average unadjusted hospitalization charges in the Philadelphia region were 158% higher than in the rest of the state (other than Pittsburgh); after adjustment, the figure was 135% higher.

 Average unadjusted hospitalization charges for amputations ($64,548) and for other procedures ($47,833) were 317% and 209% higher, respectively, than for those with no procedure ($15,494). Adjusted charges were 91% higher for those with amputations and 70% higher for those with other procedures, than for those without procedures.

 Average unadjusted hospitalization charges for those who had a medical misadventure during surgical and medical care ($99,745) were 233% higher than the charges for those who did not have medical misadventure ($29,937), but reduced to 20% higher after adjusting for the other factors.

Diabetes-related hospital charges are driven by complications, hospital misadventures, procedures, and other patient and discharge characteristics. Readmission charges are not different from initial admission charges.

Zhen-qiang Ma, MD, MPH, MS

Articles

Medicaid-Insured and Uninsured Were More Likely to Have Diabetes Emergency/Urgent Admissions

Characteristics Driving Higher Diabetes-Related Hospitalization Charges in Pennsylvania