Deintensification of Treatment With Sulfonylurea and Insulin After Severe Hypoglycemia Among Older Adults With Diabetes

Key Points Question What is the incidence of sulfonylurea or insulin treatment deintensification among older adults after an episode of severe hypoglycemia? Findings In this cohort study of 76 278 older adults with diabetes who had a hypoglycemia-associated emergency department visit or hospitalization while receiving sulfonylurea and/or insulin therapy, deintensification of the hypoglycemic treatment regimen occurred in fewer than 50% of episodes within 100 days after the event. Meaning This study’s findings suggest that a gap exists between evidence-based and current practices regarding deintensification of hypoglycemia-inducing medications among older adults with diabetes who experienced severe hypoglycemia requiring an emergency department visit or hospitalization.


Introduction
Older adults with diabetes are at higher risk of hypoglycemia compared with younger populations, 1 and this higher risk can lead to substantial morbidity and mortality. 2 Other established factors associated with hypoglycemia that may further compound this risk among older adults include frailty, multimorbidity, and longer duration of diabetes, 3 all of which are common in this population.
Notably, the development of hypoglycemia among older adults is often associated with excessively intensive diabetes treatment regimens, 4 particularly in those receiving sulfonylurea, glinide, and insulin therapies. 4,5 Care guidelines recommend individualized deintensification of hypoglycemiainducing medications in older adults with a high risk of hypoglycemia. 1 However, despite evidence of safety and efficacy, 6 multiple clinical and epidemiologic studies [7][8][9][10] have suggested that treatment deintensification remains uncommon in this population, even among those with substantial frailty and multimorbidity.
Older adults with a history of severe hypoglycemia have an especially high risk of similar future events. 5 Therefore, hypoglycemia-associated emergency department (ED) visits or hospitalizations among older adults should prompt close evaluation of the diabetes treatment regimen, and deintensification is likely reasonable in most cases. Despite the fact that hospitalization rates for hypoglycemia among older adults now exceed those for hyperglycemia (up to 105 admissions per 100 000 person-years), 11 data on patterns of hypoglycemia-associated ED visits or hospitalizations are limited, and evidence thus far suggests no substantial change in prescription fills for diabetes medications after such an event. 10 Robust data on real-world deintensification practices, including insight into the specific patient factors associated with treatment deintensification, are currently lacking.
An important step toward improving guideline-concordant care and reducing hypoglycemiaassociated morbidity and mortality among older adults is to characterize treatment deintensification practices, particularly among those with the highest risk of severe hypoglycemia. To this end, our objective was to examine deintensification of treatment with sulfonylurea and insulin, the most common hypoglycemia-inducing medications, after a hypoglycemia-associated ED visit or hospitalization among older adults with diabetes and to identify patient factors associated with treatment deintensification.

Methods
This study was approved by the institutional review board of the University of North Carolina at Chapel Hill, with a waiver of informed consent because of the use of deidentified data. The study protocol was registered with the European Network of Centres for Pharmacoepidemiology and Pharmacovigilance (registration number EUPAS37902) before accessing outcome data. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies. 12

Study Population
We analyzed medical claims and prescription drug records to obtain a random sample of 20% of nationwide fee-for-service Medicare beneficiaries. The Medicare fee-for-service database includes health insurance claims for patients with Medicare part A (inpatient services), part B (outpatient care), and part D (outpatient prescription drug) coverage. We included beneficiaries aged 65 years and older who had at least 1 hypoglycemia-associated ED visit or hospitalization between January 1, 2007, and December 31, 2017 (eFigure in the Supplement), defined using the Ginde et al 13 algorithm, with specificity greater than 99%. Consistent with previous studies, 5,13 we adapted this algorithm to include codes from the International Classification of Diseases, Tenth Revision, Clinical Modification (eTable 1 in the Supplement). We required that hypoglycemia be in the primary coding position (vs any position) to further increase the likelihood that hypoglycemia was the predominant reason for presentation to the hospital. A hypoglycemia-associated ED visit or hospitalization was the index event, and each occurrence was treated as a separate index event, even if it occurred multiple times for the same patient. If claims for ED and hospital visits were submitted on consecutive days, they were collapsed into a single encounter. Episodes were categorized by date using an ordinal variable to signify first episode vs recurrent episodes.
Participants entered the cohort on the date of the index hypoglycemia-associated event. We included individuals with at least 1 prescription for sulfonylurea, insulin, or both sulfonylurea and insulin within 6 months before the index event. We excluded individuals without at least 6 months of continuous enrollment in Medicare parts A, B, and D before the index event and patients without 100 total days of follow-up.

Outcomes
The primary outcome was deintensification of treatment with sulfonylurea and/or insulin in the 100 days after a hypoglycemia-associated ED visit or hospitalization. We defined deintensification as either (1) no prescription filled for baseline sulfonylurea and/or insulin within 100 days after the index event, (2) a prescription filled for sulfonylurea within 100 days after the index event at a lower dose than that dispensed in the last prescription before the index event, or (3) a change from long-acting sulfonylurea to glipizide within 100 days after the index event ( Table 1). The baseline (preindex event) treatment regimen was obtained using pharmacy dispensing claims data in the 100 days before a hypoglycemia-associated ED visit or hospitalization. We chose a relatively narrow period (approximately 3 months after the index event) to define treatment deintensification because a severe episode of hypoglycemia in an older adult typically prompts rapid review of the drug regimen by prescribers. The selection of 100 days (vs 90 days) was made to accommodate a 10-day period for detection of treatment deintensification among individuals who may have filled a 90-day prescription immediately before the index event.

Patient Factors
Patient factors associated with sulfonylurea and/or insulin deintensification were assessed in the 6 months before each index event. We selected the following demographic and clinical characteristics based on their potential relevance to treatment deintensification: age, sex, self-reported race and

Statistical Analysis
The demographic and clinical characteristics of Medicare beneficiaries with at least 1 hypoglycemiaassociated ED visit or hospitalization were summarized using descriptive statistics (frequencies with percentages and means with SDs). We first quantified the unadjusted 100-day incidence of sulfonylurea and/or insulin deintensification after a hypoglycemia-associated ED visit or hospitalization in the entire cohort and according to baseline receipt of hypoglycemic medications (categorized as sulfonylurea only, insulin only, and sulfonylurea and insulin). We summarized the incidence by year to examine temporal changes in treatment deintensification.
We used multivariable logistic regression models to quantify the association between individual characteristics and the odds of treatment deintensification vs no treatment deintensification after a hypoglycemia-associated ED visit or hospitalization. The final model included demographic and clinical characteristics, presence of diabetes complications and comorbidities, receipt of glucoselowering medications, receipt of other medications, and health care use and access (measured by the number of office visits). Dementia, arthritis, bladder dysfunction, hyperlipidemia, and heart failure were correlated with the Faurot frailty index (r > 0.5) 14 and removed from the final models, which had a maximum variance inflation factor of 1.5. To aid in the interpretation of insulin deintensification rates given the limitations in capturing changes in insulin dosing using claims data, we constructed a parallel analysis in an independent cohort using an index event (ie, cataract surgery) that was unlikely to impact diabetes treatment. The goal of this sensitivity analysis was to describe the background insulin deintensification rates among a sample of older adults with diabetes who were not defined by a high risk of hypoglycemia. We expected the rates of insulin deintensification in the cataract surgery cohort to be reflective of rates that may occur at random in the population and may be lower than those observed in the hypoglycemia cohort.

Study Sample
The study cohort comprised 76 278 distinct Medicare beneficiaries (eFigure in the Supplement), with a mean (SD) age of 76.6 (7.6) years. A total of 15 696 episodes were excluded from our cohort because the individuals died, representing an all-cause 100-day mortality rate of 11.5% after a hypoglycemia-associated ED visit or hospitalization (eFigure in the Supplement).

JAMA Network Open | Diabetes and Endocrinology
a Other self-identified races and ethnicities included Asian, North American Native, unspecified race or ethnicity, and unknown race or ethnicity.

Discussion
In this cohort study of an older adult population with diabetes, we found sulfonylurea and/or insulin deintensification rates of less than 50% after a hypoglycemia-associated ED visit or hospitalization, although insulin deintensification rates were likely underestimated because of the inability to capture changes in insulin dosing using claims data. Although there was a pattern of increasing sulfonylurea and/or insulin deintensification over a 10-year period, low overall rates of deintensification suggest that real-world practice may lag behind evidence that positions severe hypoglycemia as a major health and safety concern among older adults.
We found that 11.5% of the Medicare cohort died within the 100 days after hospital presentation for hypoglycemia (eFigure in the Supplement). This finding is consistent with existing knowledge that suggests an association between severe hypoglycemia and all-cause mortality, [16][17][18] and it underscores the importance of hypoglycemia avoidance in this population. Our deintensification definition did not allow us to discern whether patients who died experienced treatment deintensification before death; therefore, we had to base our analysis on survival.
Older adults with severe hypoglycemia are at a high risk of repeated events 5 ; thus, in most cases, consideration of deintensification of hypoglycemia-inducing agents is warranted after an ED visit or hospitalization for hypoglycemia. Yet, rates of sulfonylurea and/or insulin deintensification were lower than 50% in our study, and the odds of deintensification were not appreciably higher among individuals with repeated episodes (vs 1 episode) of severe hypoglycemia requiring hospital attention or among individuals older than 75 years. Our findings were similar to those of previous studies [7][8][9][10]19 that reported low rates of treatment deintensification among older adults, including adults with frailty, multimorbidity, and low life expectancy. These data suggest a need for enhanced recognition of individuals at high risk for recurrent hypoglycemic episodes as well as efforts to combat clinical inertia associated with treatment deintensification in this setting.
We found that treatment deintensification practices varied by race and ethnicity. Compared  about the appropriateness of deintensification on a case-by-case basis. Further work is necessary to understand these associations and to identify and correct health disparities that may place specific racial and ethnic subgroups at a higher risk of future hypoglycemic events.
We also found chronic kidney disease, a history of falls, and depression to be associated with significantly more frequent deintensification, regardless of baseline treatment regimen. Similar findings were observed for cerebrovascular disease and ischemic heart disease. All of these comorbidities are associated with a markedly high risk of severe hypoglycemia, 5 suggesting that health care professionals may identify and act upon high-risk clinical phenotypes, including those that would be most affected by the sequelae of hypoglycemia. Consistent with recommended practices, 1 we further observed a consistent pattern of more frequent sulfonylurea and/or insulin deintensification among older adults with higher markers of frailty. This finding is in contrast to those of a study in which insulin discontinuation among adults aged 75 to 79 years was most common in healthier individuals vs those with multimorbidity. 9 However, treatment deintensification increased only modestly with increasing frailty, so there likely remain many missed opportunities for deintensification when clinically indicated.
Notably, a lower likelihood of deintensification was found among patients receiving angiotensin-converting enzyme inhibitors and angiotensin II receptor blockers as well as statins, which suggests that patients with diabetes complications that were being managed more aggressively were less likely to experience treatment deintensification.
We observed a lower incidence of, and variability in, insulin deintensification compared with sulfonylurea deintensification. This difference may be due to decreased sensitivity in our insulin deintensification classification because of the methodologic challenges associated with capturing changes in insulin dosing in claims data. In the present study, insulin deintensification was defined as discontinuation of at least 1 type of insulin received at baseline. Although this is a standard approach used by other researchers, 6,8,19,20 it does not capture treatment deintensification in a scenario in which a prescriber advises lowering the insulin dose but does not discontinue insulin altogether. A sensitivity analysis of unadjusted insulin deintensification rates in an independent cohort that was not expected to have undergone medication changes (ie, patients after outpatient cataract surgery) found insulin deintensification rates to be approximately 7% lower. Together, these data reveal opportunities for future studies to develop and refine methods to more accurately capture insulin deintensification.

Strengths and Limitations
This study has several strengths. We used a large, nationally representative population to identify patterns in sulfonylurea and insulin deintensification and examine changes in deintensification practices over a 10-year period. Our definition of sulfonylurea deintensification included changing from a long-acting to a short-acting sulfonylurea, which allowed a more inclusive assessment of prescriber practices aimed at lowering future hypoglycemia risk. We also examined a wide range of patient factors that may be associated with deintensification, and these data can be used to inform interventions to promote evidence-based care as it pertains to minimizing hypoglycemia in older adults.
This study also has limitations. This cohort did not include patients with severe hypoglycemia who did not require hospital attention; thus, the findings may not be generalizable to individuals who experience such episodes but are treated at home. We may not have detected deintensification that occurred before death within 100 days. In addition, although individuals receiving sulfonylurea therapy are likely to have type 2 diabetes, those who receive insulin only may include individuals with type 1 diabetes for whom insulin deintensification or discontinuation may be inappropriate. However, individuals with type 1 diabetes likely make up only a small sample of the Medicare population with diabetes. In addition, we were not able to assess hemoglobin A 1c , although an association between hemoglobin A 1c and hypoglycemia-associated ED visits or hospitalizations has not been found. 10 Furthermore, even among those with high hemoglobin A 1c levels, severe hypoglycemia is a potentially life-threatening complication that requires a change in therapy in most cases, particularly among older adults for whom the short-term consequences of severe hypoglycemia substantially exceed those of hyperglycemia. Because we used pharmacy dispensing claims to define treatment deintensification, we also do not know whether deintensification was the result of a prescription change by the prescriber or the possibility that the patient did not fill the prescription. In addition, it is not possible to draw conclusions regarding the appropriateness of deintensification (or lack thereof) for the individual cases reflected in these data.

Conclusions
This cohort study found that, although the rates of treatment deintensification slowly improved between 2007 and 2017, deintensification of sulfonylurea and/or insulin therapy within 100 days after a hypoglycemia-associated ED visit or hospitalization occurred for fewer than 50% of older adults with diabetes. These findings suggest that greater efforts are needed to identify individuals at high risk of recurrent hypoglycemia and to encourage appropriate and equitable treatment deintensification practices for patients hospitalized with hypoglycemia to reduce hypoglycemiaassociated morbidity and mortality.