Primary Care Visits and Ambulatory Care Sensitive Diabetes Hospitalizations among Adult Alabama Medicaid Beneficiaries

Abstract

Purpose:

To describe patterns of care use for Alabama Medicaid adult beneficiaries with diabetes and the association between primary care utilization and Ambulatory Care Sensitive (ACS) diabetes hospitalizations.

Methods:

This retrospective cohort study analyzes Alabama Medicaid claims data from January 2010 to April 2018 for 52,549 covered adults ages 19–64 with diabetes. Individuals were characterized by demographics, comorbidities, and health care use including primary, specialty, mental health and hospital care. Characteristics of those with and without any ACS diabetes hospitalization are reported. A set of 118,758 observations was created, pairing information on primary care use in one year with ACS hospitalizations in the following year. Logistic regression analysis was used to assess the impact of primary care use on the occurrence of an ACS hospitalization.

Results:

One third of the cohort had at least one ACS diabetes hospitalization over their observed periods; hospital users tended to have multiple ACS hospitalizations. Hospital users had more comorbidities and pharmaceutical and other types of care use than those with no ACS hospitalizations. Controlling for other types of care use, comorbidities and demographics, having a primary care visit in one year was significantly associated with a reduced likelihood of ACS hospitalization in the following year (odds ratio comparing 1–2 visits versus none 0.79, 95% confidence interval 0.73–0.85).

Conclusions:

Program and population health interventions that increase access to primary care can have a beneficial effect of reducing excess inpatient hospital use for Medicaid covered adults with diabetes.

Diabetes and disease sequelae represent a significant chronic disease burden that disproportionately affects minority and low-income populations, and residents of the Southeastern United States [1, 2]. Individuals with diabetes use more medical care, have greater out-of-pocket medical costs and experience losses of productivity of over $200 billion annually [3]. Medicaid is a United States’ government program that provides health coverage to low-income families, qualified pregnant women, and some individuals with disability, with specific eligibility criteria varying by state; it is the single largest source of health coverage in the United States [4]. Diabetes prevalence is higher and complication rates are greater for the low income populations covered by Medicaid the program than for the overall population [5]. Within Medicaid programs, estimates are that adults with diabetes have annual health care costs that are triple those without diabetes [6].

One challenge facing Medicaid programs, managed care entities and other insurers is how to translate clinical knowledge about the benefits of intensive, preventive health care and diabetes self-management into interventions that can benefit low-income populations [7]. These populations have varying arrays of personal resources for diabetes-self management and varying levels of access to primary care. Primary care systems themselves may be fragmented with limited information systems that create obstacles to the delivery of practice-guideline adherent care [8].

Ambulatory care sensitive (ACS) hospitalizations, that is, hospitalizations for sets of diagnoses that panels of experts have identified as probably avoidable with timely and effective primary care [9, 10] are frequently used proxies for measuring access to primary care. The Agency for Healthcare Quality and Research has designated four such sets of ambulatory care sensitive diabetes diagnoses, and recommended their use as Prevention Quality Indicators (PQIs). These can function as screening tools to indicate areas of primary care and population health management that would benefit from quality improvement [11, 12]. It is currently unclear what the prevalence of ACS hospitalizations is among Medicaid beneficiaries with diabetes, and whether guideline-concordant primary care is associated with their reduction in this population. This study examines the frequency and pattern of occurrence of three types of ACS diabetes hospitalizations in an adult Medicaid-covered population in Alabama and their associations with the use of primary care services.

Research Design and Methods

This study was approved by the UAB Institutional Review Board. It consists of a retrospective analysis of 1/2010 – 4/2018 Medicaid paid claims data for Alabama beneficiaries between the ages of 19 and 65.

Alabama is one of the 17 states that has not yet expanded Medicaid and as such the Medicaid-covered adult population includes only parents with family incomes at or below 18% of the Federal Poverty Level and adults deemed eligible for the Federal Supplemental Security Income (SSI) program because of disability [13]. Pregnant women up to 146% of the Federal Poverty Level are also covered by Alabama Medicaid during their pregnancy, but are excluded from this study the short-term nature of their coverage and potential confounding of prenatal care and primary care use. During this time period the income-eligible and disabled Alabama Medicaid populations were enrolled in a primary care case management (PCCM) program which assigned a primary care physician to serve as a gatekeeper for access to specialty care and other services. These primary care providers were paid a supplemental monthly fee to serve in this role. As in all Medicaid programs [14], enrollees in Alabama Medicaid often experience gaps in enrollment due to changes in income or delays in annual recertification.

Population

To identify beneficiaries with diabetes, we applied the designated diagnoses used by the CMS Chronic Condition Data Warehouse project (https://www.ccwdata.org/web/guest/condition-categories) to inpatient, outpatient, and physician files to identify all claims with a diabetes diagnosis. After excluding laboratory claims (where a diagnosis could refer to a “rule-out”), we identified all unique beneficiaries and determined the date of the first claim (index date) and the number of claims with a diagnosis of diabetes for each. We then extracted all hospital and physician claims for these beneficiaries, and extracted select claims for procedures and pharmaceuticals of interest.

Outcome

The hospital claims for each individual were flagged for discharge diagnoses that included three categories of ambulatory care sensitive (ACS) diabetes hospitalizations, as defined by the Agency for Healthcare Research and Quality [11]: PQI 14 (uncontrolled diabetes), PQI 1 (diabetes short term complications such as ketoacidosis hyperosmolarity and coma), and PQI 3 (diabetes long term complications, including renal, eye, neurological and circulatory complications with a primary diagnosis of diabetes). The fourth type of ACS diabetes hospitalization defined by the Agency, lower extremity amputation, could not be identified with Alabama claims data because surgical procedures were not consistently coded on hospital claims.

Primary care

Physician claims were counted as indicating primary care use if the specialty on the claim was recorded as internal medicine, family medicine, general practitioner or nurse practitioner. They were counted as specialty care use if the specialty on the claim was recorded as nephrology, cardiology, podiatry or endocrinology, or counted as mental health care if the specialty was recorded as psychology, psychiatry or mental health clinic. We identified Hba1c testing based on procedure codes recorded on laboratory claims. Based on the National Drug Code (NDC) recorded on pharmacy claims, we identified claims for hypoglycemic, injected antidiabetics, insulin, metformin and other diabetes drugs. Both primary care use and occurrence of HbA1c tests are presented as the portion of observed months with any occurrence of claims for these services.

To identify comorbidities, we used all diagnoses recorded on claims incurred by the beneficiaries of interest over their entire observation period to count comorbidities as defined by the Charlson index [15]. The count of comorbidities was used in conjunction with age of the individual to calculate the overall Charlson comorbidity index for the observation period (https://www.mdcalc.com/charlson-comorbidity-index-cci).

Data on ACS hospitalizations and primary care were summarized by person-month. We chose to count months with this care, rather than counting the individual number of events of this care, to avoid conflating the number of individual claims filed for payment, including multiple claims for the same hospital stay, with the number of unique medical care events. This may under-represent the actual quantity of care used, but represent more adequately the use of care over time. Demographic data for each individual are stored as monthly records in the Alabama Medicaid claims data system. We linked age, race/ethnicity, county of residence and Medicaid eligibility category to the monthly files. The latest residence county over the period of observation was categorized as urban (metropolitan), small town (micropolitan), and rural based on 2013 census data compiled by the Center for Business and Economic Research at the University of Alabama (http://cber.cba.ua.edu/edata/maps/AlabamaMaps1.html). Age was also categorized based on the latest month of observation.

We excluded months where the individual was dually eligible for Medicaid and Medicare or eligible for Medicaid coverage due to pregnancy only, and months of enrollment that preceded the first paid claim with a diabetes diagnosis (i.e., before diabetes diagnosis). We aggregated the month data into one summary record for each individual. This yielded a file of 68,108 individuals. To make it more likely that individuals truly had diabetes, we excluded individuals from this file if they had less than three claims with a diabetes diagnosis, unless those claims were for a diabetes drug or a hospitalization. This yielded an analytic file of 52,549 people. In the process of aggregating months for the individual summary record, we noted periods over the observation period where the individual exited and then re-entered Medicaid enrollment. Enrollment in Medicaid for family planning services only and not for medical care coverage, was counted as a gap in enrollment.

Analysis

We used the data set of these 52,549 individuals enrolled in Medicaid between 2010 and 2018 to explore the pattern of occurrence of the three ACS hospitalizations noted above. We then compared demographic, comorbidities and health care use characteristics for those with and without occurrence of any of these ACS hospitalizations over their observed periods

In order to assess the association between primary care use and ACS hospitalizations during the following year, we created 7 sets of paired observations. For individuals observed for at least 24 months, we paired their primary care use data for months 1–12 (initial year) with their ACS hospitalization data in months 13–24 (subsequent year). We continued this pattern for individuals observed for 36, 48, 60, 72,84 and 96 months. By design, there were multiple observations for individuals observed across time; an individual who was observed for 96 months contributed data to each of the seven sets of paired years. We combined the seven sets which yielded 118,758 observations, each including a two year pairing of primary care use in the initial year and ACS hospitalization data in the subsequent year for a total of 30,643 unique individuals.

We performed unadjusted and adjusted logistic regression using the SurveyLogistic procedure in SAS 9.4 to model ACS hospitalization in the subsequent year by primary care use in the initial year, estimating robust standard errors to take into account the clustering of observations by individual. In adjusted analysis, we included demographic characteristics: age, race/ethnicity, gender, rurality of residential location, Medicaid eligibility, categorized as either related to income or to disability, and an indicator for any gaps in Medicaid enrollment in the initial year of each pair. Health status was measured by the Charlson comorbidity index [15]. Care use measures in the model included amount of primary care and HbA1c use as described above, any use of specialty care and mental health care, paid claims for insulin, metformin and other drugs, categorized separately, and any month with an ACS hospitalization in the initial year of the pair.

Results

Table 1 shows that individuals with ACS hospitalizations are slightly more likely to be African American, male, have ages between 55 and 64, and be eligible for Medicaid through disability status than those without such hospitalizations. They are observed over a longer period of time, are more likely to have had a gap in enrollment during that time, and are much more likely to have multiple comorbid conditions. There is no difference in the distribution of rural, small town and urban residents between those who did and did not have an ACS hospitalization (p=0.88).

Table 1 –

Comparison of Medicaid Recipients’ Characteristics by Diabetes-related Ambulatory Care Sensitive (ACS) Hospitalization (Alabama, 2010–2018)

Enrollee Characteristics	% Total Population N=52,549	% of Subgroup with No ACS Hospitalization N=34,645	% of Subgroup with ACS Hospitalization N=17,904	P-value*
Gender				<0.001
Male	31.0	30.3	32.3
Female	69.0	69.7	67.7
Race/Ethnicity				<0.001
Black	45.8	44.7	48.1
White	42.2	42.9	40.7
Hispanic	1.0	1.1	0.8
Other/Unknown	10.9	11.3	10.3
Age				<0.001
19–34 years	13.6	14.6	11.8
35–54 years	43.1	43.0	43.2
55–64 years	43.3	42.4	45.1
Number of months observed				<0.001
1–12 months	25.6	32.2	12.9
13–24 months	16.2	17.9	13.0
25–36 months	11.6	11.8	11.2
36–48 months	8.7	8.4	9.2
49–60 months	7.5	6.9	8.7
61–72 months	6.6	5.9	8.0
73–84 months	6.2	5.1	8.3
85–96+ months	17.5	117	28.7
Urban County	65.8	65.8	66.9	0.88
Small Town County	13.8	13.7	13.9
Rural County	20.3	20.3	20.4
Any period of eligibility based on				<0.001
Disability	74.1	70.4	81.2
Income	26.8	29.6	21.6
Any gap in enrollment over observation period
Disabled eligibility	7.1	6.3	8.4	<0.001
Income eligibility	26.1	24.4	31.5	<0.001
Number of comorbid conditions				<0.001
0	0.2	0.3	0
1	38.0	46.2	22.2
2	42.4	40.6	46.0
3 or more	19.3	12.9	31.9
Mean (SD) Charlson Comorbidity Index	5.60 (3.71)	4.80 (3.48)	6.83 (3.84)	<0.001

^**P-values calculated using chi-square tests for categorical variables or t-tests for continuous variables

Table 2 shows that over one-third of those observed experienced at least one month with an ACS hospitalization, and most of these had multiple months with hospitalizations over their observed periods. The plurality of ACS hospitalizations was for long term complications associated with diabetes; hospitalizations for uncontrolled diabetes were the next largest category, and the remaining 20% were for short term complications.

Table 2 –

Frequency of Diabetes-related Ambulatory Care Sensitive (ACS) Hospitalizations among 52,549 Medicaid Recipients with Diabetes (Alabama, 2010–2018)

Type of Hospitalization	Total number of individuals with hospitalizations* N (%)	Total number of months with hospitalizations n (%)	Mean (SD) number of months with hospitalization
Any ACS diabetes hospitalization	17,904 (34.1)	102,780 (100)	5.7 (9.4)
Uncontrolled diabetes	11,147 (21.2)	35,612 (34.6)	3.2 (5.5)
Short term diabetes complications	7,425 (14.1)	21,323 (20.7)	2.9 (4.4)
Long term diabetes complications	11,499 (21.9)	45,845 (44.6)	4.0 (6.7)

^*Individuals can have more than one type of hospitalization

Table 3 shows that individuals with ACS hospitalizations were more likely to have multiple primary care visits, HbA1c tests, mental health care, specialty visits and filled prescriptions for diabetes medications than those without such hospitalizations. Table 4 shows the results of the analysis that paired care use in one 12-month observation period with any ACS hospitalization in the following 12-month period. Without adjusting for other factors, those with more than four primary care visits in the initial year are more likely to have an ACS hospitalization in the subsequent year. Otherwise, primary care use was statistically unrelated to the likelihood of an ACS hospitalization. However, controlling for demographics, comorbidities and other types of service use, all amounts of primary care use in the initial year are associated with lower likelihoods of hospitalization in the subsequent year.

Table 3 -.

Comparison of Medicaid Recipients’ Primary Care Utilization by Diabetes-related Ambulatory Care Sensitive (ACS) Hospitalization (Alabama, 2010–2018)

Primary Care Characteristic	% Total Population N=52,549	% No ACS Hospitalization N=34,645	% With ACS Hospitalization N=17,904	P-value**
Primary care use over observed period, annualized*				<0.001
None	7.3	9.7	2.7
1–2	72.5	76.3	65.0
3–4	15.0	10.8	23.3
+ 4	5.2	3.2	9.0
Diabetes drug prescription filled				<0.001
Any type of diabetes drug	80.9	75.2	92.1
Any Metformin prescription	61.8	59.3	66.7
Any insulin prescription	40.5	26.0	68.4
Any other diabetes drug	43.5	35.8	58.3/
Hba1c test use over observed period, annualized				<0.001
None	21.9	28.3	9.6
1–2 months per year	77.6	71.5	89.4
3–4 months per year	0.4	0.2	1.0
+ 4 months per year	0	0	0
Any mental health care visit	25.3	22.6	30.6	<0.001
Any specialty visit	34.7	25.0	53.4	<0.001

^*Measured by number of months in the previous 12 months with a claim for primary care

^**P-values calculated using chi-square tests for categorical variables

Table 4 -.

Unadjusted and Adjusted Association between Medicaid Recipients’ Primary Care Use and the Likelihood of a Diabetes-related Ambulatory Care Sensitive (ACS) Hospitalization (Alabama, 2010–2018)

Variable	Any ACS Hospitalization in Subsequent Year (N= 118,463)
	Unadjusted Odds Ratio (CI)	Adjusted Odds Ratio (CI)
Primary care use in previous 12-month period*
1–2	1.002 (0.933 – 1.077)	0.786 (0.728 – 0.848)
3–4	0.948 (0.882 – 1.019)	0.711 (0.661 – 0.766)
+ 4	1.235 (1.151 – 1.325)	0.736 (0.686 – 0.79)
Demographics
Male		1.059 (1.017 – 1.103)
Black		0.941 (0.906 – 0.977)
Other		0.991 (0.925 – 1.062)
Hispanic		1.018 (0.795 – 1.303)
Age in years		0.983 (0.981 – 0.985)
Disability vs income eligible		1.01 (0.951 – 1.073)
Any gap in enrollment		0.993 (0.916 – 1.077)
Urban		1.02 (0.976 – 1.067)
Small town		0.956 (0.899 – 1.015)
Medications
Any Metformin use		1.132 (1.091 – 1.175)
Any Insulin use		2.667 (2.567 – 2.771)
Any other diabetes drug use		1.205 (1.161 – 1.251)
Charlson comorbidity index		1.105 (1.1 – 1.111)
Utilization
Any mental health service use		0.982 (0.941 – 1.025)
Any specialty care visits		1.37 (1.265 – 1.484)
Hospitalization in previous year		5.688 (5.444 – 5.944)

^*Measured by number of months in the previous 12 months with a claim for primary care

Taking primary care visits into account in the adjusted analysis, having a paid claim for a diabetes drug and having a specialty care visit are both associated with a greater likelihood of an ACS hospitalization in the subsequent year. Individuals with a higher comorbidity index and those with a hospitalization in the initial year are more likely to have an ACS hospitalization in the subsequent year. Demographic characteristics associated with a greater likelihood of a subsequent year ACS hospitalization include younger age and being male. Compared to those whose race is recorded as white and non-Hispanic in the Medicaid eligibility files, people whose race is recorded as Black had a lower likelihood of a second year ACS hospitalization.

Discussion

This analysis indicates that, among adult Alabama Medicaid beneficiaries with diabetes, hospitalization for ambulatory care sensitive diabetes diagnoses is relatively common. Contact with a primary care physician in the previous 12 months is associated with a lower likelihood of a subsequent ambulatory-care sensitive (ACS) hospitalization, when comorbidities and other factors are taken into account. This finding provides validation that the ACS hospitalizations are inversely associated with primary care visits.

Similar to our findings, ACS hospitalization rates (broadly and not just limited to those related to diabetes) have been associated with primary care in studies conducted in multiple countries, including Canada, Australia, Spain, Brazil and the United States. [16] While there is evidence of an association with primary care generally, some studies have suggested that the impact of primary care on ACS hospitalizations may be more potent for individuals of higher socioeconomic status. For example, Roos et al. found that residents in low-income neighborhoods had higher rates of ACS hospitalizations as well as higher rates of primary care visits when compared with residents of high-income neighborhoods, despite the existence of universal health care coverage. [17] Studies conducted in the United Kingdom also suggest that ACS hospitalization rates may be more reflective of socioeconomic status and patient comorbidities that of access to primary care. [18, 19] A scoping review conducted in France by Cartier et al. drew similar conclusions. [20] While there is no high-income subgroup in the current study for comparison, our findings do demonstrate that within a low-income population, primary care visits are associated with fewer ACS-hospitalizations, suggesting that primary care is preventive when income level is closer to equal.

In the current study, short-term and uncontrolled diabetes represent a majority (56%) of ACS hospitalizations. A recent study examining trends of diabetes related preventable hospitalization in the United States from 2005 to 2014 [21] found that although age-adjusted rates of preventable hospitalizations did not differ significantly over time, the distribution of ACS hospitalizations did shift, with cases of short-term complications increasing over time, constituting a majority of ACS hospitalizations, and cases of uncontrolled diabetes decreasing over time. These results differ from a previous study of ACS diabetes hospitalizations conducted in 2006 with a nationally representative sample in the U.S. [12] that identified hospitalizations for long term complications as the most common, accounting for 62% of the total, while hospitalizations for short term complications (28%) and for uncontrolled diabetes (11%) were less common. These shifts in patterns of ACS hospitalization can be used to better target interventions aimed at improving diabetes management in ambulatory settings.

Similar to previous studies [22, 23], this study showed that, in general, individuals with diabetes-specific hospitalizations are more frequent users of other health care services, including specialty care and previous hospitalizations. This is likely because their conditions are more advanced and complex, a notion further supported by the higher number of comorbidities observed for the population with hospitalizations compared to those with no ACS hospitalizations and well as the higher likelihood of being treated with insulin. Clearly, care use for diabetes does not occur in a vacuum, but in the context of care use for multiple other health problems. This has important implications for primary care management, as the literature on the adoption of the chronic care model in primary care suggests [24, 25].

After comorbidities and previous care use are taken into account, three demographic factors are associated with later diabetes ACS hospitalizations: younger age and being male are associated with a greater likelihood of hospitalization and being black instead of white or other race/ethnicity is associated with a lower likelihood of hospitalization. Rural residence is not significantly associated with the likelihood of an ACS hospitalization. Note that, because the entire population in this study are Medicaid recipients, race and rural residence are not correlated with socioeconomic status, as they are in many studies of the general population. Thus the race, age and gender factors that are associated with an ACS hospitalization in this study may be indicators of unmeasured differences in health status, or they may indicate patterned differences in personal resources that support or inhibit diabetes self-management. Such associations echo the cautions expressed by the American Diabetes Association in its current guidelines on diabetes care [8], which identify a range of factors that can create obstacles to effective diabetes self-management and optimal primary care use.

This study is subject to the limitations of all analyses of administrative data: there is very limited ability to take the nuances of individual health status into account, and limited information that would indicate lifestyle variations or differences in capacity for diabetes self-management. Claims data can indicate whether physicians were paid for seeing an individual patient, but do not provide information on the content of the interaction. Thus we are unable to determine accurately whether the primary care use described here is practice guideline-compliant. Moreover, conclusions about causal effects of primary care on ACS hospitalizations should be made with caution given the non-experimental study design. Lastly, we have examined an Alabama Medicaid population, and thus results may not generalize to the overall Medicaid population in the US.

In conclusion, findings from this study suggest that contact with primary care providers reduces the likelihood of being hospitalized for ambulatory-care sensitive conditions related to diabetes for adult Medicaid beneficiaries in Alabama. Efforts of population health managers in lowering barriers to access to primary care are an important component of improving management for this burdensome and prevalent chronic condition.

Highlights.

• ACS diabetes hospitalizations are common in adult Alabama Medicaid beneficiaries

• Primary care in past 12 months lowers likelihood of ACS hospitalization

• Younger age and being male increases likelihood of ACS hospitalization