Behavioral health conditions and potentially preventable diabetes-related hospitalizations in the United States: Findings from a national sample of commercial claims data

Objective To characterize the relationship between potentially preventable hospitalizations (PPHs) for diabetes and behavioral health conditions in commercially insured working-age persons with diabetes in the United States. Research design and methods We retrospectively analyzed medical and pharmacy claims from services rendered between 2011 and 2013 for 229,039 adults with diabetes. Diabetes PPHs were identified using the Agency for Healthcare Research and Quality’s Prevention Quality Indicators v6.0 logic. We used negative binomial-logit hurdle regression models to explore the adjusted relationships between diabetes PPHs and schizophrenia, bipolar, depression, anxiety, adjustment disorder, alcohol use disorder, and drug use disorder. Results A total of 4,521 diabetes PPHs were experienced by 3,246 of the persons in the sample. The 20.83% of persons with one or more behavioral health conditions experienced 43.62% (1,972/4,521; 95% CI 42.18%-45.07%) of all diabetes PPHs, and the 7.14% of persons with more than one diagnosed behavioral health condition experienced 24.77% (1,120/4,521; 95% CI 23.54%-26.05%) of all diabetes PPHs. After adjusting for sociodemographic and physical health covariates, patients with depression, schizophrenia, drug or alcohol use disorders, or multiple behavioral health conditions were at significantly increased risk of experiencing at least one diabetes PPH, while patients with depression, drug use disorder, or multiple behavioral health conditions were at significantly increased risk of experiencing recurring diabetes PPHs over time. Conclusions A number of behavioral health conditions are associated with diabetes PPHs, which are often preventable with timely, high-quality outpatient care. The results of this study will enable clinicians, payers, and policy-makers to better focus outpatient care interventions and resources within the population of persons with diabetes.


Supplemental File S4
Below are detailed results of two negative binomial-logit hurdle regression models examining the relationship between behavioral health conditions and potentially preventable hospitalizations for diabetes. Model 1 includes individual behavioral health conditions. Model 2 includes a count of behavioral health diagnoses.
A review of the findings for non-behavioral health variables follows the tables. The results related to the behavioral health variables are reviewed in the text of the manuscript.
Model 1: Adjusted odds ratios (aOR) and adjusted risk ratios (aRR) of potentially preventable hospitalizations for diabetes (PPH). Logit results examine findings regarding the presence or absence of any PPH, whereas the negative binomial results examine the findings regarding the increasing counts of PPHs in individuals with one or more PPH. The findings in this table are from Model 1, in which behavioral health variables are entered into the model individually.

Relationships between Diabetes PPHs and Sociodemographic Covariates
In both adjusted and unadjusted analyses, males were more likely than females to experience at least one diabetes PPH (p<0.001 for both). In unadjusted analyses males were also more likely to experience higher counts of diabetes PPHs; however, in adjusted analyses there was no gender difference in the risk of increasing counts of diabetes PPHs in persons with at least one diabetes PPH (Model 1 p=0.367; Model 2 p=0.515). In all analyses, younger age was associated with both an increased likelihood of having at least one diabetes PPH and increased counts of diabetes PPHs (p<0.001 to p=0.035). Region was also associated with diabetes PPHs in both unadjusted and adjusted analyses (p<0.001). See Supplemental File S4 for detailed results.
In unadjusted and adjusted analyses, county-level poverty was associated with an increased likelihood of having at least one diabetes PPH (p=0.007 and p=0.006, respectively).
County-level poverty was also associated with increasing counts of diabetes PPHs in unadjusted analyses (p=0.010) but this relationship was not significant in adjusted analyses (Model 1 p=0.552; Model 2 p=0.515). The risk of diabetes PPHs also varied by urban-rural category in both sets of analyses. Most notably, in adjusted analyses persons in micropolitan and non-core (i.e., rural) areas were less likely to have any diabetes PPHs, and those with at least one diabetes PPH in micropolitan areas were less likely to have increasing counts of PPHs (see Supplemental File S4).

Relationships between Diabetes PPHs and Physical Health Status Covariates
In unadjusted and adjusted analyses, chronic pain, tobacco use, chronic heart conditions, and chronic hypertension were associated with the likelihood of having at least one diabetes PPH and with increasing counts of diabetes PPHs (p<0.05 for all). In addition, in unadjusted analyses chronic renal failure, chronic cerebrovascular disease, and obesity were associated with 8 increasing counts of PPHs. They were also associated with an increased likelihood of having at least one diabetes PPH in both unadjusted and adjusted analyses (p<0.05 for all). On the other hand, these three conditions were not significantly associated with increasing counts of diabetes PPHs in those with at least one diabetes PPH in adjusted analyses (p≥0.05 for all).
In unadjusted analyses, neither asthma nor cancer were associated with an increased likelihood of having at least one diabetes PPH (p=0.186 and p=0.737, respectively), nor were they associated with increasing counts of diabetes PPHs in unadjusted (p=0.374 for both) or adjusted (p≥0.05 for all) analyses. Conversely, in adjusted analyses, the relationships between these conditions and having at least one diabetes PPH were significant (p<0.05 for all). Further, in unadjusted analyses chronic hyperlipidemia had a protective effect, significantly reducing the likelihood of having any or increasing counts of diabetes PPHs (p<0.05 for all), but in adjusted analyses only the relationship between chronic hyperlipidemia and reduced odds of having at least one PPH.